A method and system for generating an intelligent dance costume based on multi-modal motion analysis

Abstract

The application provides a kind of intelligent dance clothing generation method and system based on multi-modal action analysis. Through the inertial sensor array and multi-view visual system, dance biomechanics data is collected, features are extracted using spatio-temporal graph convolution network, style semantics are analyzed using CLIP model, and design drawings are generated through diffusion model and integrated with physical constraint optimization. The 3D model is bound with a virtual digital person to simulate dynamic effects, and production instructions containing fabric, pattern and process parameters are output. The system includes data acquisition, action analysis and other modules, realizing full-process intelligentization. This method breaks through the traditional limitations, integrates biomechanics data and artistic style semantics, improves the clothing tear resistance, style matching accuracy and shortens the design cycle, realizes the automatic design and production of dance clothing, improves the dynamic adaptability and artistic expression of clothing, and provides an efficient solution for dance clothing customization.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and clothing design technology, specifically relating to an intelligent dance costume generation method and system based on multimodal motion analysis. Background Technology

[0002] With the development of stage arts, dance costumes must simultaneously meet the dual requirements of dynamic functionality (such as range of motion and tear resistance) and artistic expression (such as stylistic semantics and pattern details). Traditional design methods rely on manual experience, resulting in low efficiency, poor accuracy, and difficulty in quantitative verification. In the field of stage dance, the core value of dance costumes always revolves around "movement adaptation"—all style designs, fabric selections, and pattern structures must ultimately serve the smooth execution of dance movements.

[0003] This means that the rational design logic for dance costumes should be "deriving design parameters from the movement requirements": first, clarify the dynamic characteristics of the movement (such as joint range of motion and stress intensity), then determine the ease of the garment's cut, fabric elasticity, and craftsmanship details accordingly, and finally incorporate stylistic elements. However, the current industry's technical system has long been trapped in the mindset of "forward design," which is seriously out of touch with the actual needs of "movement-driven design," leading to a series of intractable technical pain points.

[0004] Existing technologies attempt to incorporate digital tools such as 3D scanning and computer vision, but they have not yet broken through the framework of "forward design." The core deficiency lies in the inability to derive design parameters from actions:

[0005] 1. Lack of dynamic data, no basis for reverse design: Existing technology can only obtain the static dimensions of the human body through static 3D scanning or capture the contours of movements through single-modal vision, and cannot simultaneously collect dynamic biomechanical parameters in dance movements;

[0006] 2. Style-action disconnect, no mechanism for reverse adjustment: Existing style semantic parsing technology can only convert text descriptions into static elements such as patterns and colors, and cannot combine action requirements to optimize the style structure in reverse.

[0007] 3. Production parameter gaps and lack of reverse implementation link: Even if the pattern is generated through forward design, the existing technology cannot reverse map the parameters derived from the motion to the production process.

[0008] In summary, the core bottleneck in the current intelligent design of dance costumes is not the lack of digital tools, but the lack of a technical system that "takes the needs of dance movements as a starting point and reverse-engineers the parameters of the entire design and production chain." Summary of the Invention

[0009] This invention aims to provide an intelligent dance costume generation method and system based on multimodal motion analysis. Through innovative technologies such as multimodal data fusion, spatiotemporal feature extraction, and physical constraint optimization, it captures dynamic biomechanical data of movements and drives style analysis, pattern generation, and production parameter matching in reverse, thus solving the above-mentioned technical bottlenecks and realizing the intelligent design and production of dance costumes.

[0010] In a first aspect, this invention proposes an intelligent dance costume generation method based on multimodal motion analysis, which includes the following steps:

[0011] S1: Multimodal biomechanical data of dance movements are collected synchronously through an inertial sensor array and a multi-view visual motion capture system. The biomechanical data includes human key point detection, skeleton modeling, joint angular velocity and limb movement trajectory.

[0012] S2: The spatiotemporal features of the dance movements are extracted using a spatiotemporal graph convolutional network. Inertial sensor data and visual point cloud data are fused through a node feature aggregation layer to construct a three-dimensional graph feature that integrates spatiotemporal information. The data of each dimension are processed by Min-Max normalization and stored in tensor format.

[0013] S3: The CLIP-based dance style semantic parser performs cross-modal alignment of user-input text prompts and generates a basis feature tensor for stylized clothing by combining it with a local clothing image library.

[0014] S4: Based on the stylized base feature tensor and the 3D atlas features fused with spatiotemporal features, a diffusion model is invoked to generate preliminary costume design drawings. The spatiotemporal features and diffusion model feature layers are fused through a dynamic attention mechanism. Based on a physical simulation constraint model, the cutting lines, elastic fiber distribution, and tear resistance parameters are optimized to generate an initial version of the dance costume plan and 3D modeling file, incorporating dance style.

[0015] S5: Bind the generated 3D modeling file to the skeleton of the virtual digital human, simulate the dynamic wearing effect based on the physics engine, and output a production instruction file with associated fabric library material parameters, pattern library size specifications, and process parameter library production constraints.

[0016] Preferably, step S2 specifically includes:

[0017] S201: Employs a dual-stream spatiotemporal alignment architecture to process inertial sensor data and visual point cloud data; Inertial data stream processing includes Butterworth low-pass filtering of joint quaternion data acquired by the inertial sensor array to retain effective biomechanical signals in the 0-10Hz range, and converting raw sensor data including acceleration and angular velocity into 6-DOF joint poses including 3-axis position and 3-axis rotation; Visual data stream processing includes 3D point cloud registration and skeletal topology reconstruction, using the ICP algorithm to stitch multi-view point clouds into a complete human body model for point cloud registration, and generating a skeletal topology structure containing 26 key nodes based on a graph optimization algorithm;

[0018] S202: The time delay difference τ of multimodal data is calculated using the generalized cross-correlation algorithm, and a spatiotemporal coordinate system transformation model is established. The formula for calculating the time delay difference R(τ) between inertial data and visual data is as follows: R(τ) =argmax τ ∑ t X inertial ( t )· X visual ( t + τ ), where argmax τ This represents the time delay τ that causes the subsequent summation operation to reach its maximum value; t is the time series index, corresponding to the discrete time point of data acquisition. X inertial and X visual These represent the normalized inertial and visual data, respectively.

[0019] Preferably, the spatiotemporal features in step S2 include human keypoint detection, skeletal topology modeling, joint torque distribution, motion trajectory curvature, and center of gravity offset. The dimensions of the three-dimensional map features are T×V×C, where T is the time frame, V is the number of joint vertices, and C is the number of channels.

[0020] Preferably, the cross-modal alignment in step S3 includes:

[0021] Step 301: Encode the text prompts into semantic vectors using the CLIP model, and encode the local clothing image into a visual vector;

[0022] Step 302: Calculate the cosine similarity between the semantic vector and the visual vector, and select images with a similarity > 0.85 as stylized basis feature tensors;

[0023] Step 303: Inject the stylized basis feature tensor into the diffusion model to control the texture details of the generated image.

[0024] Preferably, step S4 specifically includes:

[0025] Step 401: Input the three-dimensional map features into a spatiotemporal pooling encoder, compress them into a two-dimensional style conditional vector with dimension 1×D through temporal average pooling and spatial max pooling; perform spatial deformation fitting on the stylized basis feature tensor through deformable convolutional layers, and output a semantic feature vector that matches the dimension of the spatiotemporal conditional vector; the dimension of the basis feature tensor is H×W×D, where H is height, W is width, and D is depth;

[0026] Step 402: During the diffusion model denoising process, physical constraints are injected. Based on the biomechanical data of dance movements, the functional constraint parameters of clothing are derived through the physical simulation pre-calculation module. This includes: calculating the stress distribution matrix of the cutting line based on the differential equation of the movement trajectory curvature and joint angular velocity; deriving the linear fiber density map by simulating the stretching threshold of fabrics of different materials through finite element analysis; and encoding the obtained constraint parameters into physical guidance vectors, which are then concatenated with spatiotemporal condition vectors and style semantic vectors to form a multi-condition joint embedding with a dimension of 1×3D.

[0027] Step 403: The condition generation of the diffusion model includes: in the cross-attention layer of the diffusion model UNet, the physical constraint parameters are backpropagated to the latent space by the projection gradient descent method;

[0028] Step 404: Lightweight simulation verification: Perform rapid physical verification on the generated clothing design drawing, bind the mesh model of the generated clothing 3D model file to the virtual digital human skeleton, and simulate the fabric deformation under high impact based on the mass-spring system; if the deformation of key parts exceeds the preset threshold, a local regeneration mechanism is triggered: mask repair is performed on the defect area in the potential space of the diffusion model.

[0029] Preferably, step S5 specifically includes:

[0030] Step 501: Import the 3D modeling file of the generated clothing into the virtual digital human skeleton system and bind key stress points;

[0031] Step 502: Drive the movement of the digital human based on motion capture data, and calculate the displacement of the garment vertex and the fabric stretch rate in real time;

[0032] Step 503: If the garment deformation in the simulation results exceeds the safety threshold, adjust the size specifications in the associated pattern library, match the material with an elastic modulus ≥ 50MPa in the fabric library, and write the optimized pattern parameters into the stylized base feature tensor to achieve closed-loop iteration.

[0033] More preferably, the security threshold in step 503 includes:

[0034] Under high-speed rotation, the side seam stretch rate is ≤ 80% of the fabric's breaking elongation.

[0035] During a high jump, the groin deformation displacement should be ≤ 90% of the pattern redundancy.

[0036] When the threshold is exceeded, the pattern ease is automatically adjusted by 1-5cm, and the fabric with an elastic modulus ≥50MPa is matched.

[0037] Secondly, embodiments of the present invention provide an intelligent dance costume generation system based on multimodal motion analysis, comprising:

[0038] The data acquisition module is configured to synchronously acquire biomechanical data of dance movements through an inertial sensor array and a visual motion capture system. The biomechanical data includes human key point detection, skeleton modeling, joint angular velocity, and limb movement trajectory.

[0039] The motion parsing module is configured to extract the spatiotemporal features of the dance motion using a spatiotemporal graph convolutional network, fuse inertial sensor data and visual point cloud data through a node feature aggregation layer, construct a three-dimensional graph feature that integrates spatiotemporal information, and process the data of each dimension through Min-Max normalization and store it in tensor format.

[0040] The multimodal generation module is configured to perform cross-modal alignment of user-input text prompts by a CLIP-based dance style semantic parser, and generate a basis feature tensor for stylized clothing by combining it with a local clothing image library.

[0041] The physical simulation module is configured to use the stylized base feature tensor and the 3D atlas features fused with spatiotemporal features. It calls the diffusion model to generate a preliminary costume design drawing, fuses the spatiotemporal features with the diffusion model feature layer through a dynamic attention mechanism, optimizes the cutting lines, elastic fiber distribution and tear resistance parameters based on the physical simulation constraint model, and generates an initial version of the dance costume plan drawing and 3D modeling file that combines dance style.

[0042] The production interface module is configured to bind the generated 3D modeling file to the skeleton of the virtual digital human, simulate the dynamic wearing effect based on the physics engine, and output a production instruction file that associates the material parameters of the fabric library, the size specifications of the pattern library, and the production constraints of the process parameter library.

[0043] More preferably, the parameter mapping rules in the production interface module include:

[0044] Set up an elastic modulus mapping, select fabrics with an elastic modulus ≥ 50 MPa based on the physical simulation results, and match the spandex blend material number in the fabric library.

[0045] Set the pattern size mapping, and adjust the pattern size redundancy corresponding to the stress hot spot area according to the deformation gradient ratio. For every 10% increase in deformation gradient, the ease increases by 1cm.

[0046] Set process parameter mapping, mapping tear resistance requirements to a sewing thread density of ≥12 stitches / cm, and using a four-thread overlocking process at the seams.

[0047] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any of the implementations of the first aspect.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] (1) Dynamic modeling driven by multimodal data: Dance motion data is collected synchronously by inertial sensor array and multi-view vision system. Combined with dual-stream spatiotemporal alignment architecture, the spatiotemporal unification of biomechanical features (joint angular velocity, trajectory curvature) and visual point cloud is achieved. Compared with single-modal scheme, motion analysis error is significantly reduced.

[0050] (2) Co-optimization of style semantics and physical constraints: The CLIP model is used to construct a dance style semantic parser, which transforms text prompts into stylized base features. At the same time, biomechanical features are injected into the diffusion model through a dynamic attention mechanism to achieve the joint generation of "motion functionality" (such as tear resistance) and "artistic expression" (such as pattern details).

[0051] (3) Full-chain digital production closed loop: By simulating the dynamic wearing effect through the physical engine, the production instructions related to fabric parameters, pattern specifications and process constraints are automatically output. Compared with the traditional process, the pattern making efficiency is improved and the dynamic adaptability of clothing is significantly improved. Attached Figure Description

[0052] The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments and, together with the description, serve to explain the principles of the invention. Other embodiments and many anticipated advantages of the embodiments will be readily recognized as they become better understood through reference to the following detailed description. Elements in the drawings are not necessarily to scale. The same reference numerals refer to corresponding similar parts.

[0053] Figure 1 This is a schematic diagram of the overall process of the intelligent dance costume generation method based on multimodal motion analysis according to an embodiment of the present invention;

[0054] Figure 2 This is a diagram of a dual-stream spatiotemporal alignment architecture according to an embodiment of the present invention;

[0055] Figure 3 This is a schematic diagram illustrating the process of generating a style base according to an embodiment of the present invention;

[0056] Figure 4 This is a schematic diagram of the process for generating clothing after cross-modal alignment according to an embodiment of the present invention;

[0057] Figure 5 This is a schematic diagram of the architecture of an intelligent dance costume generation system based on multimodal motion analysis, as an embodiment of the present invention.

[0058] Figure 6 This is a schematic diagram of the structure of a computer device suitable for implementing electronic devices according to embodiments of the present invention. Detailed Implementation

[0059] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0060] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0061] The present invention aims to disclose an intelligent dance costume generation method and system based on multimodal motion analysis. By integrating biomechanical data and artistic style semantics, it realizes the automated design and production of dance costumes, and improves the dynamic adaptability and artistic expression of the costumes.

[0062] In a first aspect, embodiments of the present invention disclose an intelligent dance costume generation method based on multimodal motion analysis, such as... Figure 1 As shown, the method includes the following steps:

[0063] S1: Multimodal biomechanical data of dance movements are collected synchronously through an inertial sensor array and a multi-view visual motion capture system. The biomechanical data includes human key point detection, skeleton modeling, joint angular velocity and limb movement trajectory.

[0064] S2: The spatiotemporal features of the dance movements are extracted using a spatiotemporal graph convolutional network. Inertial sensor data and visual point cloud data are fused through a node feature aggregation layer to construct a three-dimensional graph feature that integrates spatiotemporal information. The data of each dimension are processed by Min-Max normalization and stored in tensor format.

[0065] Specifically, the spatiotemporal features mentioned in step S2 include human key point detection, skeletal topology modeling, joint torque distribution, motion trajectory curvature and center of gravity offset. The dimensions of the three-dimensional map features are T×V×C, where T is the time frame, V is the number of joint vertices, and C is the number of channels.

[0066] The spatiotemporal features of the dance movements are extracted using a spatiotemporal graph convolutional network (ST-GCN). Inertial sensing data and visual point cloud data are fused through a node feature aggregation layer to construct a three-dimensional graph feature that integrates spatiotemporal information.

[0067] like Figure 2 As shown, step S2 specifically includes:

[0068] S201: Employs a dual-stream spatiotemporal alignment architecture to process inertial sensor data and visual point cloud data;

[0069] The dual-stream spatiotemporal alignment architecture is a synchronization and fusion technology framework designed for multimodal data (inertial sensor data + visual point cloud data). Its core objectives are to address the following issues: First, temporal alignment, eliminating timestamp discrepancies caused by differences in sensor sampling frequencies (e.g., 200Hz for inertial sensors vs. 120Hz for visual cameras). Second, spatial alignment, mapping different coordinate systems (local coordinate system of the inertial sensor vs. global coordinate system of the visual system) to a unified reference system. Third, data integrity, repairing data loss caused by occlusion or sensor failure. A data repair mechanism is constructed, using cubic spline interpolation to repair missing data, ensuring spatiotemporal continuity.

[0070] Processing the inertial data stream: This includes performing Butterworth low-pass filtering on the joint quaternion data acquired by the inertial sensor array to retain the effective biomechanical signals in the 0-10Hz range, and converting the raw sensor data, including acceleration and angular velocity, into 6-DOF joint poses, including 3-axis position and 3-axis rotation.

[0071] Visual data stream processing includes 3D point cloud registration and skeletal topology reconstruction. The ICP (Iterative Closest Point) algorithm is used to stitch multi-view point clouds into a complete human body model for point cloud registration. A skeletal topology structure containing 26 key nodes is generated based on a graph optimization algorithm.

[0072] S202: The time delay difference τ of multimodal data is calculated using a generalized cross-correlation algorithm, and a spatiotemporal coordinate system transformation model is established to map different coordinate systems (the local coordinate system of the inertial sensor vs. the global coordinate system of the vision system) to a unified reference system. The formula for calculating the time delay difference R(τ) between inertial and visual data is: R(τ) =argmax τ ∑ t X inertial ( t )· X visual ( t + τ ), where argmax τ This represents the time delay τ that causes the subsequent summation operation to reach its maximum value; t is the time series index, corresponding to the discrete time point of data acquisition. X inertial and X visual These represent the normalized inertial and visual data, respectively.

[0073] S3: The CLIP-based dance style semantic parser performs cross-modal alignment of user-input text prompts and generates a basis feature tensor for stylized clothing by combining it with a local clothing image library.

[0074] Specifically, such as Figure 3 As shown, the cross-modal alignment in step S3 includes:

[0075] Step 301: Encode the text prompts into semantic vectors using the CLIP model, and encode the local clothing image into a visual vector;

[0076] Step 302: Calculate the cosine similarity between the semantic vector and the visual vector, and select images with a similarity > 0.85 as stylized basis feature tensors;

[0077] Step 303: Inject the stylized basis feature tensor into the diffusion model to control the texture details of the generated image.

[0078] S4: Based on the stylized base feature tensor and the 3D atlas features fused with spatiotemporal features, a preliminary costume design drawing is generated by calling the stable diffusion model. The spatiotemporal features and the diffusion model feature layer are fused through a dynamic attention mechanism. Based on the physical simulation constraint model, the cutting lines, elastic fiber distribution, and tear resistance parameters are optimized to generate an initial version of the dance costume plan drawing and 3D modeling file that combines dance style.

[0079] like Figure 4As shown, step S4 generates dance costume design drawings using a diffusion model guided by physical constraints. The dynamic attention mechanism includes:

[0080] Step 401: Input the 3D map features into a spatiotemporal pooling encoder, compress them into a 2D style conditional vector with dimension 1×D through temporal average pooling and spatial max pooling; perform spatial deformation fitting on the stylized basis feature tensor through a deformable convolutional layer, and output a semantic feature vector that matches the dimension of the spatiotemporal conditional vector; the dimension of the basis feature tensor is H×W×D, where H is height, W is width, and D is depth;

[0081] Step 402: During the diffusion model denoising process, physical constraint conditions are injected. Based on the biomechanical data of the dance movements, the functional constraint parameters of the clothing are derived through the physical simulation pre-calculation module. This includes: calculating the stress distribution matrix of the cutting line based on the differential equation of the curvature of the movement trajectory and the joint angular velocity; deriving the linear fiber density map by simulating the stretching threshold of different fabric materials through finite element analysis; encoding the obtained constraint parameters into a physical guidance vector (Physics Guidance Embedding), and concatenating it with the spatiotemporal condition vector and the style semantic vector to form a multi-condition joint embedding with a dimension of 1×3D; where, physical constraint condition injection refers to injecting physical constraint conditions based on the biomechanical data of the dance movements (joint torque distribution, center of gravity offset).

[0082] Preferably, step 402 includes a Hadamard product operation, which satisfies the following conditions: the number of channels in the attention weight matrix is ​​consistent with the number of channels in the feature map of the diffusion model; the normalization range of the weight matrix is ​​[0,1], and the weight distribution is constrained by the Softmax function; weights are dynamically injected in the 20th-40th sampling steps of the diffusion model to control the curvature and elasticity distribution of the garment cutting lines.

[0083] The Hadamardi layer is a key bridge connecting multimodal data and generative models. By weighting elements one by one, it achieves efficient integration of the biomechanical characteristics of dance movements, the semantics of artistic style, and the physical properties of clothing. This ensures both the movement functionality of the clothing and the aesthetic requirements of stage performance, reflecting the organic combination of "data-driven" and "rule-constrained" in intelligent design.

[0084] Step 403: The condition generation of the diffusion model includes: in the cross-attention layer of the diffusion model UNet, the physical constraint parameters are backpropagated to the latent space by the projection gradient descent method;

[0085] Specifically, the conditional generation of the diffusion model uses an improved Stable Diffusion architecture, adding a multi-conditional guided branch to its UNet's Cross-Attention Layer: Query vector: style cue word embedding from the diffusion model's text encoder; Key-Value pair: from the multi-conditional joint embedding, mapped to the attention space through a linear projection layer; during the diffusion sampling process, the physical constraint parameters are backpropagated to the latent space through Projected Gradient Descent, dynamically adjusting the generated results to meet the requirements of tear resistance and motion adaptability.

[0086] Step 404: Lightweight simulation verification: This refers to the rapid physical verification of the generated clothing design drawings. The mesh model of the generated clothing 3D model file is bound to the virtual digital human skeleton, and the fabric deformation under high impact is simulated based on the mass-spring system. If the deformation of key parts exceeds the preset threshold, a local regeneration mechanism is triggered: masking repair (inpainting) is performed on the defect area in the potential space of the diffusion model.

[0087] In one specific embodiment, if the deformation of a key component exceeds a preset threshold, including a knee elongation rate > 15%, a local regeneration mechanism is triggered. It should be noted that the indicators for the deformation of key components can be selected and adjusted according to actual needs, and are not specifically limited here.

[0088] S5: Bind the generated 3D modeling file to the skeleton of the virtual digital human, simulate the dynamic wearing effect based on the physics engine, and output a production instruction file with associated fabric library material parameters, pattern library size specifications, and process parameter library production constraints.

[0089] Specifically, the steps of the S5 physics engine simulation include:

[0090] Step 501: Import the 3D modeling file of the generated clothing into the virtual digital human skeleton system and bind key stress points;

[0091] Step 502: Drive the movement of the digital human based on motion capture data, and calculate the displacement of the garment vertex and the fabric stretch rate in real time;

[0092] Step 503: If the garment deformation in the simulation results exceeds the safety threshold, adjust the size specifications in the associated pattern library, match the material with an elastic modulus ≥ 50MPa in the fabric library, and write the optimized pattern parameters into the stylized base feature tensor to achieve closed-loop iteration.

[0093] Furthermore, the safety thresholds mentioned in step 503 include: the side seam stretch rate under high-speed rotation is ≤ 80% of the fabric breaking elongation; the crotch deformation displacement under large jump is ≤ 90% of the pattern redundancy; when the thresholds are exceeded, the pattern ease is automatically adjusted in increments of 1-5cm, and a fabric with an elastic modulus ≥ 50MPa is matched.

[0094] Further reference Figure 5 As an implementation of the methods shown in the above figures, this application provides an embodiment of an intelligent dance costume generation system based on multimodal motion analysis. This system embodiment is similar to... Figure 1 Corresponding to the method embodiments shown, the system can be specifically applied to various electronic devices.

[0095] Secondly, embodiments of the present invention also disclose an intelligent dance costume generation system based on multimodal motion analysis, such as... Figure 5 As shown, the system includes: a data acquisition module 51, an action analysis module 52, a multimodal generation module 53, a physical simulation module 54, and a production interface module 55.

[0096] In one specific embodiment, the data acquisition module 51 is configured to synchronously acquire dance movement biomechanical data through an inertial sensor array and a visual motion capture system. The biomechanical data includes human key point detection, skeleton modeling, joint angular velocity, and limb movement trajectory. The motion parsing module 52 is configured to extract the spatiotemporal features of the dance movement using a spatiotemporal graph convolutional network, fuse inertial sensor data and visual point cloud data through a node feature aggregation layer to construct a three-dimensional graph feature that integrates spatiotemporal information, and perform Min-Max normalization on the data of each dimension and store it in tensor format.

[0097] The multimodal generation module 53 is configured to perform cross-modal alignment of user-input text prompts using a CLIP-based dance style semantic parser, and generate a base feature tensor for stylized clothing by combining it with a local clothing image library; the physical simulation module 54 is configured to generate a preliminary clothing design drawing by calling a diffusion model based on the stylized base feature tensor and the 3D atlas features fused with spatiotemporal features, fuse the spatiotemporal features and the feature layer of the diffusion model through a dynamic attention mechanism, optimize the cutting lines, elastic fiber distribution and tear resistance parameters based on a physical simulation constraint model, and generate an initial version of a dance costume plan drawing and 3D modeling file that combines dance style;

[0098] The production interface module 55 is configured to bind the generated 3D modeling file to the skeleton of the virtual digital human, simulate the dynamic wearing effect based on the physics engine, and output a production instruction file that associates the material parameters of the fabric library, the size specifications of the pattern library, and the production constraints of the process parameter library.

[0099] More preferably, the parameter mapping rules in the production interface module 55 include:

[0100] Set up an elastic modulus mapping, select fabrics with an elastic modulus ≥ 50 MPa based on the physical simulation results, and match the spandex blend material number in the fabric library.

[0101] Set the pattern size mapping, and adjust the pattern size redundancy corresponding to the stress hot spot area according to the deformation gradient ratio. For every 10% increase in deformation gradient, the ease increases by 1cm.

[0102] Set process parameter mapping, mapping tear resistance requirements to a sewing thread density of ≥12 stitches / cm, and using a four-thread overlocking process at the seams.

[0103] Among them, in the data acquisition module 51: the inertial sensor array is deployed on the shoulder, elbow, hip and knee joints of the dancer, with a sampling frequency ≥200Hz and an angle measurement error <0.5°; the visual motion capture device includes 6 infrared cameras with a resolution of 3840×2160 and a frame rate of 120fps, used to generate three-dimensional point cloud data of limb trajectories.

[0104] The functions and methods of the above modules correspond to each other, and will not be repeated here.

[0105] This invention discloses an intelligent dance costume generation method and system based on multimodal motion analysis. By capturing dynamic biomechanical data of the movements, it reverse-drives style analysis, pattern generation, and production parameter matching. This is a key direction to break through the existing technical difficulties and meet the actual needs of the industry, and has important technological innovation value and industrial application significance.

[0106] In one specific embodiment, the specific implementation of the present invention is roughly as follows:

[0107] 1. System Architecture and Hardware Deployment

[0108] The intelligent dance costume generation system of the present invention adopts a distributed architecture and mainly includes:

[0109] Data Acquisition Layer:

[0110] Inertial sensor array: The Xsens MVN Awinda motion capture kit is selected, which includes 17 inertial measurement units (IMUs), a sampling frequency of 200Hz, and a measurement range of ±16g acceleration and ±2000° / s angular velocity, for real-time acquisition of 6-DOF pose data of human joints.

[0111] Multi-view vision system: Four Intel RealSense D435i depth cameras (1280×720 resolution, 120fps) are deployed in a rectangular arrangement in a 10m×8m acquisition space with a height of 2.5m to acquire 3D point cloud data of the human body.

[0112] Computational processing layer:

[0113] Server configuration: NVIDIA A100 GPU×4, 512GB RAM, Intel Xeon Platinum 8380 CPU, running Ubuntu 20.04 operating system.

[0114] Software stack: PyTorch 2.0 deep learning framework, CUDA 11.8 for accelerated computing, OpenCV 4.7 for computer vision processing, and Bullet Physics Engine for physics simulation.

[0115] 2. Multimodal data acquisition and preprocessing

[0116] 2.1 Data Synchronization Acquisition

[0117] Dance movement capture:

[0118] Professional dancers wear inertial sensor arrays and perform typical dance movements (such as ballet fouetté turns and modern dance leaps) within the visual system's coverage area, continuously collecting data for 30 seconds.

[0119] Spacetime synchronization mechanism:

[0120] A combination of hardware triggering and software calibration is used to achieve sub-millisecond synchronization between the inertial sensor and the camera via a Sync Pulse Generator. The data delay difference τ is then calculated using the Generalized Cross-Correlation Algorithm (GCC-PHAT).

[0121] τ = argmax_τ ∫ X_inertial(f)·X_visual (f)·|X_visual(f)|² df

[0122] Where argmax_τ represents the time delay τ for maximizing the integral result; f is the frequency variable, corresponding to the frequency dimension after the Fourier transform of the data; X_inertial and X_visual are the Fourier transforms of the inertial and visual data, respectively. Indicates conjugate.

[0123] 2.2 Data Preprocessing

[0124] Inertial data processing:

[0125] A Butterworth low-pass filter (4th order, cutoff frequency 10Hz) is used to remove motion noise;

[0126] The raw sensor data is converted into a joint rotation matrix using the quaternion method, and the joint angular velocity and angular acceleration are calculated.

[0127] Visual data processing:

[0128] Multi-view point cloud registration is achieved based on the Iterative Closest Point (ICP) algorithm, with an error threshold set at 0.01m.

[0129] The solvePnP algorithm from OpenCV is used for camera pose estimation to construct a world coordinate system;

[0130] Human body contours are extracted by segmenting point clouds based on a skeletal topology model (26 key nodes).

[0131] 3. Spatiotemporal Feature Extraction and Fusion

[0132] 3.1 Spatiotemporal Graph Convolutional Network (ST-GCN)

[0133] Network architecture:

[0134] A 10-layer ST-GCN is used, which includes spatial graph convolutional layers and temporal convolutional layers. The input dimension is T×V×C (T=600 frames, V=26 joints, C=128 channels).

[0135] Feature extraction:

[0136] Spatial features: Capturing the structural relationships between joints, such as the linkage between the shoulder and elbow joints;

[0137] Temporal characteristics: Extract the temporal changes of the action, such as the angular velocity change curve during the "whip spin" process.

[0138] 3.2 Multimodal Feature Fusion

[0139] Node feature aggregation layer:

[0140] Weighted fusion of inertial features (joint angular velocity) and visual features (point cloud curvature):

[0141] F_fused = α·F_inertial + (1-α)·F_visual

[0142] Where α is the adaptive weight, with a value range of [0,1], which is obtained through training; F_inertial is the inertial feature vector; F_visual is the visual feature vector; and F_fused is the fused feature vector.

[0143] Normalization process:

[0144] Min-Max normalization is applied to the fused features, mapping the feature values ​​to the [0,1] interval:

[0145] F_normalized = (F - F_min) / (F_max - F_min)

[0146] Where F is the original feature value, F_min is the minimum feature value, F_max is the maximum feature value, and F_normalized is the normalized feature value.

[0147] 4. Semantic Analysis of Dance Styles

[0148] 4.1 Application of CLIP Model

[0149] Cross-modal alignment:

[0150] The text prompts entered by the user (such as "Dunhuang Dance - Flying Apsaras Style") are encoded into a 512-dimensional semantic vector using the CLIP model;

[0151] Feature extraction was performed on a local clothing image library (5000 samples, covering 24 dance styles) to generate a visual vector library.

[0152] Similarity calculation:

[0153] Calculate the cosine similarity between the semantic vector and the visual vector, and filter images with a similarity > 0.85:

[0154] similarity = cosine_similarity(text_embedding, image_embedding)

[0155] Where text_embedding is the encoded vector of the text prompt, image_embedding is the encoded vector of the clothing image, and similarity is the cosine similarity value between the two.

[0156] 4.2 Stylized Basis Feature Generation

[0157] Style transfer is performed on the selected images, and a stylized basis feature tensor with dimensions H×W×D is generated through a deformable convolutional network (DCN).

[0158] 5. Diffusion generation under physical constraints

[0159] 5.1 Construction of Condition Vectors

[0160] Spatiotemporal condition vector:

[0161] The 3D map features are compressed into a 1×512 dimensional vector by using temporal average pooling and spatial max pooling.

[0162] Physical constraint calculation:

[0163] Based on the curvature of the motion trajectory and the angular velocity of the joints, the stress distribution matrix of the cutting line is calculated using differential equations:

[0164] σ = E·ε = E·(ΔL / )

[0165] Where E is the elastic modulus, ε is the strain, and ΔL is the tensile length. The original length.

[0166] Finite element analysis (FEA) is used to simulate the stretching behavior of fabrics and a safety threshold is set (e.g., elongation at break of spandex blended fabric ≥ 450%).

[0167] 5.2 Optimization of Diffusion Model

[0168] Multi-condition joint embedding:

[0169] The spatiotemporal condition vector, style semantic vector, and physical guidance vector are concatenated into a 1×1536-dimensional embedding vector.

[0170] Constraint injection mechanism:

[0171] In sampling steps 25-40 of the Stable Diffusion model's UNet, the physical constraints are backpropagated to the latent space using projective gradient descent.

[0172] z_t+1 = z_t - α· _z L(θ, z_t, y, c)

[0173] Where z represents the latent variable, z_t represents the latent variable at step t of the diffusion model, z_t+1 represents the latent variable after the update at step t+1, θ represents the network parameters of the diffusion model, and α represents the learning rate. _z represents the gradient with respect to the latent variable z, L represents the loss function, y represents the physical constraint parameters, and c represents the conditional embedding vector.

[0174] 6. Dynamic Simulation and Production Instruction Generation

[0175] 6.1 Physical Simulation Verification

[0176] Digital human binding:

[0177] The generated 3D clothing model is bound to the Mixamo standard skeleton, with key stress points (such as the shoulders and knees) connected using spring dampers.

[0178] Dynamic testing:

[0179] The digital human's movement is driven by the collected motion data, and the Bullet Physics Engine is used to calculate the apex displacement of the clothing and the fabric stretch rate.

[0180] 6.2 Parameter Mapping and Command Generation

[0181] Elastic modulus mapping:

[0182] When the side seam stretch rate is detected to be greater than 80% of the fabric breaking elongation, an spandex blended fabric with an elastic modulus ≥ 50 MPa (such as No. FB-2358) will be automatically matched.

[0183] Pattern Adjustment Rules:

[0184] In stress hotspot areas, the loosening amount is adjusted according to the deformation gradient. For every 10% increase in the deformation gradient, the loosening amount increases by 1 cm.

[0185] Process parameter matching:

[0186] When the tear resistance requirement is ≥8N / cm², the automatic four-thread overlock process is used, and the sewing thread density is ≥12 stitches / cm.

[0187] 6.3 Closed-loop optimization

[0188] If the simulation results are not satisfactory, the optimized template parameters are written into the stylized basis feature tensor to trigger a new round of generation, forming a closed-loop iteration.

[0189] 7. Implementation Cases

[0190] 7.1 Data Acquisition Phase

[0191] Data was collected on ballet dancers performing a fouetté turn. An inertial sensor recorded a maximum hip joint angular velocity of 198° / s, and a vision system captured a maximum torso twist angle of 127°.

[0192] 7.2 Garment Production Stage

[0193] When a user inputs the text "classical ballet - white tulle skirt", the system selects the 15 most similar images and generates stylized base features.

[0194] Physical simulations show that the hem of a conventionally designed skirt stretches by 23% (exceeding the safety threshold of 15%). The system automatically adjusts the skirt width by +3cm and increases the elastic fiber density by 20%.

[0195] 7.3 Production Output Stage

[0196] The generated production instruction file includes: fabric parameters (elastic modulus 55MPa, weight 120g / m²), pattern specifications (bust + 2cm, waist + 1cm), and process requirements (key seams use five-line overlock).

[0197] Through the above implementation methods, the present invention realizes intelligent management of the entire process of dance costumes, from motion data acquisition and style semantic analysis to physical attribute optimization, providing a data-driven innovative solution for stage costume design.

[0198] This invention fills the technological gap in intelligent design of dance costumes, realizing full-process automation from motion data acquisition and artistic style analysis to physical property optimization, and providing data and algorithm support for large-scale personalized customization of stage costumes.

[0199] The following is for reference. Figure 6 It shows a schematic diagram of the structure of a computer device 600 suitable for implementing an electronic device according to embodiments of the present invention. Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0200] like Figure 6 As shown, the computer device 600 includes a central processing unit (CPU) 601 and a graphics processing unit (GPU) 602, which can perform various appropriate actions and processes according to programs stored in read-only memory (ROM) 603 or programs loaded from storage portion 609 into random access memory (RAM) 604. The RAM 604 also stores various programs and data required for the operation of the device 600. The CPU 601, GPU 602, ROM 603, and RAM 604 are interconnected via a bus 605. An input / output (I / O) interface 606 is also connected to the bus 605.

[0201] The following components are connected to I / O interface 606: an input section 607 including a keyboard, mouse, etc.; an output section 608 including an LCD, speakers, etc.; a storage section 609 including a hard disk, etc.; and a communication section 610 including a network interface card, such as a LAN card or modem. The communication section 610 performs communication processing via a network such as the Internet. A drive 611 may also be connected to I / O interface 606 as needed. A removable medium 612, such as a hard disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 611 as needed so that computer programs read from it can be installed into storage section 609 as needed.

[0202] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 610, and / or installed from removable medium 612. When the computer program is executed by central processing unit (CPU) 601 and graphics processing unit (GPU) 602, the functions defined in the methods of this invention are performed.

[0203] It should be noted that the computer-readable medium described in this invention can be a computer-readable signal medium, a computer-readable medium, or any combination thereof. A computer-readable medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor device, or any combination thereof. More specific examples of a computer-readable medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution device, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than a computer-readable medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution device, apparatus, or apparatus. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0204] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0205] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using dedicated hardware-based devices that perform the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0206] The modules described in the embodiments of the present invention can be implemented in software or hardware. The described modules can also be located in a processor.

[0207] In another aspect, the present invention also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to perform the methods and steps described in the first aspect of the embodiments of the present invention.

[0208] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention is not limited to the specific combination of the above-described technical features, but also includes other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this invention.

Claims

1. A method for generating an intelligent dance costume based on multi-modal motion analysis, characterized in that, The method includes the following steps: S1: Multimodal biomechanical data of dance movements are collected synchronously through an inertial sensor array and a multi-view visual motion capture system. The biomechanical data includes human key point detection, skeleton modeling, joint angular velocity and limb movement trajectory. S2: The spatiotemporal features of the dance movements are extracted using a spatiotemporal graph convolutional network. Inertial sensor data and visual point cloud data are fused through a node feature aggregation layer to construct a three-dimensional graph feature that integrates spatiotemporal information. The data of each dimension are processed by Min-Max normalization and stored in tensor format. The spatiotemporal features include human keypoint detection, skeleton topology modeling, joint torque distribution, movement trajectory curvature and center of gravity offset. The dimension of the three-dimensional graph feature is T×V×C, where T is the time frame, V is the number of joint vertices, and C is the number of channels. S3: The CLIP-based dance style semantic parser performs cross-modal alignment of user-input text prompts and generates a basis feature tensor for stylized clothing by combining it with a local clothing image library. S4: Based on the stylized base feature tensor and the 3D atlas features fused with spatiotemporal features, the diffusion model is called to generate a preliminary costume design drawing. The spatiotemporal features and the feature layer of the diffusion model are fused through a dynamic attention mechanism. Based on the physical simulation constraint model, the cutting lines, elastic fiber distribution and tear resistance parameters are optimized to generate an initial version of the dance costume plan drawing and 3D modeling file that combines dance style. as well as S5: Bind the generated 3D modeling file to the skeleton of the virtual digital human, simulate the dynamic wearing effect based on the physics engine, and output a production instruction file with associated fabric library material parameters, pattern library size specifications and process parameter library production constraints; Step S4 specifically includes: Step 401: Inputting the three-dimensional map features into a spatiotemporal pooling encoder, compressing them into a two-dimensional style conditional vector with dimension 1×D through temporal average pooling and spatial max pooling; performing spatial deformation fitting on the stylized basis feature tensor through a deformable convolutional layer, and outputting a semantic feature vector that matches the dimension of the spatiotemporal conditional vector; the dimension of the basis feature tensor is H×W×D, where H is height, W is width, and D is depth; Step 402: During the diffusion model denoising process, physical constraints are injected. Based on the biomechanical data of dance movements, the functional constraint parameters of clothing are derived through the physical simulation pre-calculation module. This includes: calculating the stress distribution matrix of the cutting line based on the differential equation of the movement trajectory curvature and joint angular velocity; deriving the linear fiber density map by simulating the stretching threshold of fabrics of different materials through finite element analysis; and encoding the obtained constraint parameters into physical guidance vectors, which are then concatenated with spatiotemporal condition vectors and style semantic vectors to form a multi-condition joint embedding with a dimension of 1×3D. Step 403: The condition generation of the diffusion model includes: in the cross-attention layer of the diffusion model UNet, the physical constraint parameters are backpropagated to the latent space by the projection gradient descent method; Step 404: Lightweight simulation verification: Perform rapid physical verification on the generated clothing design drawing, bind the mesh model of the generated clothing 3D model file to the virtual digital human skeleton, and simulate the fabric deformation under high impact based on the mass-spring system; if the deformation of key parts exceeds the preset threshold, a local regeneration mechanism is triggered: mask repair is performed on the defect area in the potential space of the diffusion model. 2.The intelligent dance garment generation method based on multi-modal motion parsing according to claim 1, wherein, Step S2 specifically includes: S201: Employs a dual-stream spatiotemporal alignment architecture to process inertial sensor data and visual point cloud data; Processing the inertial data stream: This includes performing Butterworth low-pass filtering on the joint quaternion data acquired by the inertial sensor array to retain the effective biomechanical signals in the 0-10Hz range, and converting the raw sensor data, including acceleration and angular velocity, into 6-DOF joint poses, including 3-axis position and 3-axis rotation. Visual data stream processing includes 3D point cloud registration and skeletal topology reconstruction. The ICP algorithm is used to stitch multi-view point clouds into a complete human body model for point cloud registration. A skeletal topology structure containing 26 key nodes is generated based on a graph optimization algorithm. S202: The time delay difference τ of multimodal data is calculated using the generalized cross-correlation algorithm, and a spatiotemporal coordinate system transformation model is established. The formula for calculating the time delay difference R(τ) between inertial data and visual data is as follows: R(τ) = argmax τ ∑ t x inertial ( t )· x visual ( t + τ ) where argmax τ denotes finding the time delay difference τ value that maximizes the result of the subsequent summation operation; t is the time series index, corresponding to the discrete time points of data collection; X inertial and X visual denote normalized inertial and visual data, respectively. 3.The intelligent dance garment generation method based on multi-modal motion parsing according to claim 1, wherein, The cross-modal alignment mentioned in step S3 includes: Step 301: Encode the text prompts into semantic vectors using the CLIP model, and encode the local clothing image into a visual vector; Step 302: Calculate the cosine similarity between the semantic vector and the visual vector, and select images with a similarity > 0.85 as stylized basis feature tensors; Step 303: Inject the stylized basis feature tensor into the diffusion model to control the texture details of the generated image. 4.The intelligent dance garment generation method based on multi-modal motion parsing according to claim 1, wherein, Step S5 specifically includes: Step 501: Import the 3D modeling file of the generated clothing into the virtual digital human skeleton system and bind key stress points; Step 502: Drive the movement of the digital human based on motion capture data, and calculate the displacement of the garment vertex and the fabric stretch rate in real time; Step 503: If the garment deformation in the simulation results exceeds the safety threshold, adjust the size specifications in the associated pattern library, match the material with an elastic modulus ≥ 50MPa in the fabric library, and write the optimized pattern parameters into the stylized base feature tensor to achieve closed-loop iteration. 5.The intelligent dance garment generation method based on multi-modal motion parsing according to claim 4, wherein, The security threshold mentioned in step 503 includes: Under high-speed rotation, the side seam stretch rate is ≤ 80% of the fabric's breaking elongation. During a high jump, the groin deformation displacement should be ≤ 90% of the pattern redundancy. When the threshold is exceeded, the pattern ease is automatically adjusted by 1-5cm, and the fabric with an elastic modulus ≥50MPa is matched.

6. An intelligent dance wear generating system based on multi-modal motion parsing, characterized by, For implementing the method as described in any one of claims 1-5, comprising: The data acquisition module is configured to synchronously acquire biomechanical data of dance movements through an inertial sensor array and a visual motion capture system. The biomechanical data includes human key point detection, skeleton modeling, joint angular velocity, and limb movement trajectory. The motion parsing module is configured to extract the spatiotemporal features of the dance motion using a spatiotemporal graph convolutional network, fuse inertial sensor data and visual point cloud data through a node feature aggregation layer, construct a three-dimensional graph feature that integrates spatiotemporal information, and process the data of each dimension through Min-Max normalization and store it in tensor format. The multimodal generation module is configured to perform cross-modal alignment of user-input text prompts by a CLIP-based dance style semantic parser, and generate a basis feature tensor for stylized clothing by combining it with a local clothing image library. The physical simulation module is configured to use the stylized base feature tensor and the 3D atlas features fused with spatiotemporal features. It calls the diffusion model to generate a preliminary costume design drawing, fuses the spatiotemporal features with the diffusion model feature layer through a dynamic attention mechanism, optimizes the cutting lines, elastic fiber distribution and tear resistance parameters based on the physical simulation constraint model, and generates an initial version of the dance costume plan drawing and 3D modeling file that combines dance style. The production interface module is configured to bind the generated 3D modeling file to the skeleton of the virtual digital human, simulate the dynamic wearing effect based on the physics engine, and output a production instruction file that associates the material parameters of the fabric library, the size specifications of the pattern library, and the production constraints of the process parameter library.

7. The multi-modal motion resolved based smart dance wear generating system according to claim 6, wherein, The parameter mapping rules in the production interface module include: Set up an elastic modulus mapping, select fabrics with an elastic modulus ≥ 50 MPa based on the physical simulation results, and match the spandex blend material number in the fabric library. Set the pattern size mapping, and adjust the pattern size redundancy corresponding to the stress hot spot area according to the deformation gradient ratio. For every 10% increase in deformation gradient, the ease increases by 1cm. Set process parameter mapping, mapping tear resistance requirements to a sewing thread density of ≥12 stitches / cm, and using a four-thread overlocking process at the seams.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 5.

Images