Intelligent interactive costume design system and method based on data model

Abstract

The invention relates to the technical field of costume design, and discloses an intelligent interactive costume design system and method based on a data model. The system comprises an interaction module, a human body modeling module, a demand intelligent analysis module, an intelligent design generation module, a fabric matching module, a virtual fitting module and a scheme optimization module. The interaction module obtains PLY format human body data of a user and costume design requirements and receives modification feedback of the user, and the human body modeling module generates a 3D human body model, human body size parameters and posture feature tags. The demand intelligent analysis module converts a costume design demand into a structured demand document containing a style tag and a version constraint parameter in combination with a posture feature tag, the intelligent design generation module outputs an optimal costume design scheme through a VAE generation model, and the fabric matching module matches candidate fabrics based on the optimal costume design scheme. The virtual fitting module realizes dynamic fitting simulation, and the scheme optimization module iteratively optimizes the optimal costume design scheme to meet user requirements. The personalized costume design efficiency of the user is improved.

Description

A data model-based intelligent interactive clothing design system and method Technical Field

[0001] This invention relates to the field of clothing design technology, and specifically to an intelligent interactive clothing design system and method based on a data model. Background Technology

[0002] With the upgrading of consumption, traditional standardized clothing can no longer meet users' dual demands for fit and uniqueness. However, the current design and customization model relies heavily on the experience of designers, which has significant bottlenecks: manual measurement has low accuracy, which can easily lead to mismatch between the subsequent pattern and the human body shape; the process of hand-drawn design sketches, manual pattern making and sample garment production is cumbersome and has a long design cycle, making it difficult to meet the demand for rapid customization; designers need to communicate with users multiple times to understand their unstructured needs, but it is difficult to accurately combine them with human body characteristics, resulting in a large deviation between the design and the user's expectations; moreover, the high proportion of manual processes and the large amount of waste in sample garment production drive up customization costs and restrict the large-scale promotion of the industry.

[0003] While existing technologies have made some improvements to address the aforementioned issues, they have not overcome the core pain points: some systems introduce 3D scanning equipment to generate basic 3D human body models, but fail to effectively extract key human body size parameters and body shape feature labels, thus failing to provide accurate adaptation basis for design; some systems offer pseudo-customization functions that only allow users to fine-tune the appearance of clothing, without involving intelligent optimization of pattern parameters; some systems attempt to integrate natural language input functions, but due to a lack of mature processing technology, manual intervention is still required to interpret the requirements; moreover, there is a lack of a complete closed loop of design, feedback, and optimization. After users submit modification opinions, the pattern-making process must be restarted, resulting in low iteration efficiency and low user satisfaction. In summary, current technologies suffer from inaccurate utilization of human body data, unintelligent conversion of personalized needs, low design and optimization efficiency, and a lack of a closed loop for user interaction, failing to simultaneously achieve personalized customization, high-efficiency design, and high user satisfaction. Therefore, an intelligent interactive clothing design system based on data models is urgently needed to solve these problems. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent interactive clothing design system and method based on a data model, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by this invention is: an intelligent interactive clothing design system based on a data model, comprising an interaction module, a human body modeling module, a demand intelligent analysis module, an intelligent design generation module, a fabric matching module, a virtual try-on module, and a solution optimization module; the interaction module acquires the user's human body data and clothing design requirements, displays the virtual try-on effect of the clothing, and receives user feedback for modifications; the human body modeling module generates a 3D human body model based on the human body data and extracts the human body size parameters and body shape feature tags from the 3D human body model; the demand intelligent analysis module is used to combine the body shape feature tags to transform the clothing design requirements into a system containing... The system includes a structured requirements document containing style tags and pattern constraint parameters; an intelligent design generation module that inputs a 3D human body model and the structured requirements document into a VAE generation model to generate an optimal clothing design scheme; a fabric matching module that constructs a fabric database and matches candidate fabrics from the database with the optimal clothing design scheme, outputting the fabric information of the corresponding candidate fabrics; a virtual try-on module that receives the optimal clothing design scheme and fabric information, achieving dynamic fitting simulation between the 3D clothing model and the 3D human body model to generate a virtual try-on effect; and a scheme optimization module that optimizes the optimal clothing design scheme based on user feedback until it meets user needs.

[0006] Preferably, the interaction module includes a data receiving unit, an effect display unit, and a feedback acquisition unit. The data receiving unit acquires the user's human body data through a 3D scanning device. This 3D scanning device generates point cloud data of the human body surface by emitting coded structured light and receiving reflected light signals, and automatically encapsulates it into a PLY format point cloud file for storage. It also has a built-in PLY file verification module to exclude damaged or invalid PLY format point cloud files by detecting point cloud density and coordinate integrity. The data receiving unit also acquires the user's clothing design requirements described in natural language through a text input box. The text input box integrates natural language auxiliary input functionality and a syntax correction module. The natural language auxiliary input functionality provides real-time association recommendations based on a preset clothing style keyword library, which is a vocabulary set of clothing style types. The syntax correction module performs semantic correction. The system automatically prompts and corrects ambiguous statements. The effect display unit uses the Unity real-time rendering engine to construct a 3D image showcasing the virtual try-on effect of clothing. It also supports multi-dimensional viewing of the virtual try-on effect, including one-click switching of commonly used perspectives, 0%-200% zoom in on specific clothing areas, and lighting mode switching, restoring the visual effects of the virtual try-on effect in different scenarios. The feedback acquisition unit provides a graphical annotation tool, which includes a quick part location module and a convenient parameter adjustment module. The quick part location module has built-in preset labels for key clothing parts; users can click on the corresponding label to automatically locate the corresponding part to be modified in the 3D image of the virtual try-on effect. The convenient parameter adjustment module supports both numerical input and slider adjustment modes, allowing users to directly input the adjustment parameters for the part to be modified or drag the slider from the current adjustment parameter to the target adjustment parameter.

[0007] Preferably, the system includes a point cloud preprocessing unit, a 3D reconstruction unit, and a feature extraction unit. The point cloud preprocessing unit sequentially uses a statistical filtering algorithm to remove noise points from the PLY format human body data, performs coordinate registration based on the alignment algorithm of the human torso's central axis to unify the 3D coordinate system, and performs point cloud smoothing using the moving least squares method to obtain standardized human body point cloud data. The 3D reconstruction unit, based on the standardized human body point cloud data, constructs a closed triangular mesh model using a Poisson surface reconstruction algorithm with a depth value of 8. Then, combined with human anatomical features, it adjusts and optimizes the mesh curvature of key areas, including but not limited to the shoulders, waist, and hips, to ensure that the model fits the real human body shape and generates a 3D human body model. The feature extraction unit extracts human body size parameters such as chest circumference, waist circumference, hip circumference, shoulder width, back length, sleeve length, and upper body length from the 3D human body model using a distance and perimeter calculation algorithm based on the vertex coordinates of the 3D human body model. By comparing with preset body shape feature parameter thresholds, it determines body shape type labels, shoulder shape labels, and waist feature labels.

[0008] Preferably, the method for obtaining the structured requirement document is as follows: A BERT-base-chinese model trained on a corpus in the clothing domain is used to perform entity recognition on natural language clothing design requirements. The clothing domain corpus contains several clothing requirement texts labeled with style entities and pattern entities. The BERT-base-chinese model accurately extracts style-related and pattern-related words from the natural language clothing design requirements, forming a requirement keyword set containing style keywords and pattern keywords. Based on this, a pre-set semantic mapping table is used to associate the requirement keyword set with body shape feature labels, and pre-set rules are invoked. The engine processes the associated set of requirement keywords and body shape feature tags to generate a structured requirement document containing style tags and pattern constraint parameters. The semantic mapping table predefines the correspondence between style keywords, pattern keywords, and each body shape feature tag. The core logic of the rule engine is based on a pre-built clothing knowledge graph. The matching logic of the clothing knowledge graph is clear: for different types of body shape feature tags, when they form a combination matching relationship with the corresponding pattern keywords, a corresponding quantitative adjustment rule for the pattern parameter is preset. That is, each matching combination of body shape feature tags and pattern keywords corresponds to a specific numerical adjustment logic for a pattern parameter.

[0009] Preferably, the process of obtaining the optimal clothing design scheme is as follows: The 3D human body model is converted into a feature vector; seven human body size parameters corresponding to the 3D human body model are extracted: bust, waist, hip, shoulder width, back length, sleeve length, and upper thigh length; these seven human body size parameters are then normalized using the Z-score normalization algorithm, and finally processed using Python. The NumPy library is used to construct a 1×7 dimensional human feature vector. Simultaneously, the structured requirements document is transformed into a parameter vector. Style tags in the structured requirements document are One-Hot encoded to generate a 1×m dimensional style sub-vector (m being the number of clothing style types). Pattern constraint parameters are directly extracted to generate a 1×n dimensional pattern sub-vector (n being the number of core pattern parameter fields in the pattern constraint parameters). The pattern sub-vector and style sub-vector are concatenated to form a 1×(n+m) dimensional parameter vector. Weights are then assigned to the feature vector and parameter vector, and matrix multiplication is used to fuse them into a fixed-dimensional fusion vector. This fusion vector is input into a VAE generation model to generate multiple candidate clothing design schemes. The 3D clothing model of each candidate design scheme is output in OBJ format, the pattern parameter table is output in JSON format, and the style feature vector is output as a 1×m dimensional vector. The VAE generation model is based on several sets of feature vectors and parameters... The model is trained using a historical dataset that links numerical vectors and labeled clothing design schemes. The VAE generative model's encoder contains a 3-layer fully connected network, and the decoder contains a 3-layer deconvolutional network. The cosine similarity between the style feature vectors of each candidate clothing design scheme and the standard vectors of style tags in the structured requirements document is calculated using the cosine function of the SciPy library, with a value ranging from 0 to 1. The candidate scheme pattern parameter table and the pattern constraint parameters in the structured requirements document are then converted into vectors of the same dimension. The difference value is calculated using the Euclidean distance formula, and then normalized to obtain the pattern fit degree in the 0-1 interval. Finally, the cosine similarity and pattern fit degree are weighted at 50% each, and the matching score of each candidate clothing design scheme is calculated by weighting the cosine similarity and pattern fit degree at 50%. The candidate clothing design scheme with the highest matching score is selected as the optimal clothing design scheme. Each candidate clothing design scheme includes a pattern parameter table, a 3D clothing model, and a style feature vector.

[0010] Preferably, the fabric database uses a MySQL relational database to store the physical and visual properties of each fabric, and each fabric corresponds to a unique fabric identifier. Among them, the physical properties include gram weight, elastic modulus, and air permeability, and the visual properties include Pantone color number and texture type; the method for obtaining the fabric information of the candidate fabric is: performing vector mapping between the style feature vector of the optimal clothing design plan and the fabric visual properties. Among them, the Pantone color number is converted into a 3D color number vector through RGB color values, and the texture type is encoded into a k-dimensional type vector through One-Hot encoding (k is the number of texture types, k < m). After splicing the type vector and the color number vector into a visual property vector, the cosine similarity algorithm is used to calculate the similarity value between the style feature vector and the visual property vector. At the same time, the adaptability with the physical properties is judged according to the preset adaptation rules for the pattern types. The adaptation rules are the quantitative adaptation thresholds of the physical properties of the fabrics corresponding to each pattern type, that is, each pattern type forms a corresponding relationship with the specific numerical range of at least one physical property of the fabric, and the minimum or maximum threshold requirements that the fabric required for this pattern type needs to meet in the corresponding physical property are clarified. Comparing the pattern type in the pattern parameter table of the optimal clothing design plan with the preset adaptation rules, the fabric that meets all the threshold requirements is determined to be qualified for physical property adaptation. Accordingly, the weighted summation algorithm is used to calculate the comprehensive matching score of the fabric. The weighted summation algorithm is: fabric comprehensive matching score = similarity value × 40% + adaptability score × 60%. Among them, if the physical property adaptation is qualified, the adaptability score is 1 point, otherwise, the adaptability score is 0 point; all fabrics are sorted in descending order according to the fabric comprehensive matching score, and the fabric with the highest fabric comprehensive matching score is selected as the candidate fabric. The fabric information includes the fabric identifier, physical properties, and visual properties of the candidate fabric.

[0011] Preferably, the virtual try-on module aligns the 3D clothing model of the optimal clothing design plan with the 3D human body model through the ICP algorithm, and the alignment error is controlled within ≤0.3mm. At the same time, based on the pattern parameter table of the optimal clothing design plan, for body parts including but not limited to shoulders, waist, and hips, differential vertex weights are assigned through the grid vertex weight adjustment technology to achieve the static fitting of the 3D clothing model and the 3D human body model. The PBR technology is used to convert the Pantone color number of the candidate fabric into RGB color parameters and the texture type into a texture map with a resolution of ≥1024px. Accordingly, the PBD physical simulation engine is called, combined with the physical properties of the candidate fabric, to simulate the dynamic fold deformation of the clothing under different human postures, and generate a 360-degree rotatable three-dimensional clothing virtual try-on effect.

[0012] Preferably, the process of optimizing the clothing design scheme is as follows: using the YOLOv8 object detection algorithm to identify the parts that need to be modified as marked by the user, and extracting the adjustment parameters from the user's modification feedback of the parts that need to be modified, mapping the parts that need to be modified and the adjustment parameters to the corresponding fields in the structured requirements document, and generating an updated structured requirements document; regenerating the parameter vector based on the updated structured requirements document, and fusing it with the feature vector of the original 3D human body model to form a new fusion vector, inputting it into the VAE generation model to regenerate candidate clothing design schemes and selecting the optimal clothing design scheme.

[0013] Due to the adoption of the above technical solutions, the technological advancements achieved by this invention compared to existing technologies are as follows: The primary technological advancement of this invention lies in achieving a leap from subjective human judgment to intelligent, precise, and structured analysis of clothing design requirements. Existing technologies rely on manual analysis of natural language requirements, which is prone to distortion and difficult to correlate with human body features. This invention accurately extracts requirement keywords through a pre-trained BERT-based-Chinese model, establishes a quantitative correlation logic between keywords and body features based on a clothing knowledge graph and rule engine, and ultimately outputs a structured requirement document, solving the core pain points of poor compatibility between requirements and human body features and low structuring in existing technologies. Secondly, it achieves a revolution in design scheme generation, shifting from experience-driven manual drawing to data model-driven automatic optimization. Existing AI-assisted design tools mostly generate… This invention addresses the limitations of traditional 2D sketches requiring manual selection. It employs a VAE (Visual Architecture Engine) model and vector fusion technology, fusing standardized human feature vectors and requirement parameter vectors through an attention mechanism. This automatically generates multiple candidate solutions and selects the optimal one through quantitative scoring, fundamentally changing the current situation where existing technologies rely on designer experience, have low generation efficiency, and poor solution adaptability. Finally, it achieves a closed loop of solution iteration, moving from repetitive redesign to parameterized intelligent and rapid optimization. Traditional solution modifications require redrawing sketches and are prone to parameter errors. This invention uses an object detection algorithm to automatically identify user-annotated modification areas and parameters, converting them into adjustment instructions to update the structured requirement document. The optimized solution is then quickly output through the original generation process, significantly improving personalized customization efficiency and solving the problems of long iteration cycles and parameter transmission errors in existing technologies.

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0015] Figure 1 is a schematic diagram of the system functional modules of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1, as shown in Figure 1, is an intelligent interactive clothing design system based on a data model, including an interaction module, a human body modeling module, a demand intelligent analysis module, an intelligent design generation module, a fabric matching module, a virtual try-on module, and a solution optimization module.

[0018] The interactive module acquires the user's human body data and clothing design requirements, displays the virtual try-on effect of the clothing, and receives user feedback for modifications. The human body modeling module generates a 3D human body model based on the human body data and extracts the human body size parameters and body shape feature tags from the 3D human body model. The intelligent requirement parsing module combines the body shape feature tags to transform the clothing design requirements into a structured requirement document containing style tags and pattern constraint parameters. The intelligent design generation module inputs the 3D human body model and the structured requirement document into the VAE generation model to generate the optimal clothing design scheme. The fabric matching module builds a fabric database and matches the corresponding candidate fabrics in the fabric database according to the optimal clothing design scheme, outputting the fabric information of the corresponding candidate fabrics. The virtual try-on module receives the optimal clothing design scheme and fabric information, realizes dynamic fitting simulation between the 3D clothing model and the 3D human body model, and generates the virtual try-on effect of the clothing. The scheme optimization module optimizes the optimal clothing design scheme based on user feedback until it meets the user's requirements.

[0019] Furthermore, the working principle of this invention is illustrated below using a personalized dress customization scenario for individual users as an example: The data receiving unit of the system's interaction module is equipped with a structured light 3D scanning device, the effect display unit builds a 3D display environment based on the Unity real-time rendering engine, and the feedback acquisition unit has a built-in graphical annotation tool; the human body modeling module, the intelligent demand analysis module, the intelligent design generation module, the fabric matching module, the virtual try-on module, and the solution optimization module are all deployed on a server running Python 3.9, TensorFlow 2.8, and SciPy 1.8; the fabric database associated with the fabric matching module uses a MySQL relational database to store fabric data containing physical attributes, visual attributes, and unique fabric identifiers; simultaneously, the intelligent demand analysis module loads a BERT-base-chinese model trained on clothing domain corpus, pre-sets a semantic mapping table containing the correspondence between style keywords, pattern keywords, and body feature labels, and a rule engine built based on a clothing knowledge graph; the intelligent design generation module loads a pre-trained VAE generation model, and the virtual try-on module pre-sets parameter configuration files for the ICP algorithm, the PBD physics simulation engine, and the PBR rendering technology.

[0020] When the interaction module acquires the user's human body data and clothing design requirements, the user initiates a full-body scan using a structured light 3D scanning device through the data receiving unit, generating human body point cloud data in PLY format. The PLY file verification module of the data receiving unit checks the point cloud density and the integrity of the x / y / z three-dimensional coordinates. After confirming that there are no errors, the human body data is stored. At the same time, the user inputs the clothing design requirements "French retro style dress, hoping to make the waist look slimmer and cover the hips" in natural language through the text input box. The natural language auxiliary input function of the text input box recommends options such as "French retro - waist-cinching style" and "French retro - A-line" based on the clothing style keyword library. The syntax correction module does not detect semantic ambiguity, and finally confirms and stores the clothing design requirements.

[0021] After receiving human body data in PLY format from the interaction module, the human body modeling module first uses a statistical filtering algorithm to remove noise points from the human body point cloud data, retaining valid points with a confidence level of ≥95%. Then, an alignment algorithm based on the human torso's central axis is used to complete coordinate registration to unify the 3D coordinate system. Next, the moving least squares method is used for point cloud smoothing, outputting standardized human body point cloud data. Following this, the 3D reconstruction unit uses the standardized human body point cloud data to construct a closed triangular mesh model using a Poisson surface reconstruction algorithm with a depth value of 8. The mesh curvature of key areas such as the shoulders, waist, and hips is then adjusted based on human anatomical features. The system is adjusted and optimized to generate a 3D human body model that fits the user's real body shape. Finally, the feature extraction unit extracts human body size parameters such as chest circumference, waist circumference, hip circumference, shoulder width, back length, sleeve length, and upper thigh length by using a distance and perimeter calculation algorithm based on the vertex coordinates of the 3D human body model. It compares these parameters with preset body shape feature parameter thresholds to determine the user's body shape feature tags as "pear-shaped body (hip circumference > chest circumference by more than 5cm), normal shoulder type (shoulder slope angle 15°-25°), and narrow waist (waist circumference / height < 0.35)". The 3D human body model, human body size parameters, and body shape feature tags are then transmitted to the demand intelligent analysis module and the intelligent design generation module.

[0022] After receiving the clothing design requirements from the interaction module and the body feature tags from the human body modeling module, the intelligent requirement parsing module first calls the BERT-base-chinese model to perform entity recognition on "French retro style dress, hoping to make the waist look slimmer and hide the hips," extracting a set of requirement keywords such as "French retro, dress, waist-cinching, A-line skirt." Then, it uses a pre-set semantic mapping table to associate the set of requirement keywords with body feature tags such as "pear-shaped body tag," "normal shoulder type tag," and "slim waist tag." Finally, it calls the rule engine to correct and constrain the association results based on the matching logic of the clothing knowledge graph (e.g., pear-shaped body tag + A-line skirt style keyword corresponds to "hip width increase of 3cm," slim waist tag + waist-cinching style keyword corresponds to "waist circumference reduction of 2cm," etc.), and finally generates a structured requirement document containing "style tag: French retro; style constraint parameters: shoulder width increase of 1cm, waist circumference reduction of 2cm, hip width increase of 3cm, sleeve length of 50cm, and skirt A-line skirt," and transmits it to the intelligent design generation module.

[0023] After receiving the 3D human body model from the human body modeling module and the structured requirement document from the intelligent requirement analysis module, the intelligent design generation module first performs vector transformation: the human body size parameters of the 3D human body model are normalized using the Z-score normalization algorithm, and a 1×7-dimensional human body feature vector is constructed using the Python NumPy library. The "French retro" style tag from the structured requirement document is encoded using One-Hot to generate a 1×10-dimensional style sub-vector (covering 10 mainstream clothing styles). The pattern constraint parameters are extracted to generate a 1×5-dimensional pattern sub-vector, which is then concatenated into a 1×15-dimensional parameter vector. Next, vector fusion is performed: the human body feature vector is weighted at 60% and the parameter vector at 40% using an attention mechanism, and then fused into a 1×20-dimensional fusion vector through matrix multiplication. Finally, model generation and selection are performed: the fusion vector is input into the VAE generation model to generate three candidate clothing design schemes. Each candidate clothing design scheme contains a 3D clothing model in OBJ format, a pattern parameter table in JSON format, and a 1×10-dimensional style feature vector. The style feature vector of each candidate clothing design scheme is calculated. The cosine similarity between the quantity and the standard vector of "French retro" was calculated. The cosine similarity values ​​of the three candidate clothing design schemes were 0.92, 0.85, and 0.88, respectively. The pattern parameter table and the pattern constraint parameters of the structured requirements document were converted into vectors of the same dimension. The normalized pattern fit was calculated. The pattern fit values ​​of the three candidate clothing design schemes were 0.95, 0.88, and 0.91, respectively. The matching score was calculated with a 50% weighting for style similarity and a 50% weighting for pattern fit. The matching scores were 0.935, 0.865, and 0.895, respectively. The candidate scheme with the highest matching score was selected as the optimal clothing design scheme (including a pattern parameter table with a 1cm shoulder width allowance, a 2cm waist narrowing, and a 3cm hip allowance, etc., and a 3D clothing model in OBJ format with a French retro style). This was then transferred to the fabric matching module and the virtual try-on module.

[0024] After receiving the optimal garment design scheme from the intelligent design generation module, the fabric matching module first extracts the style feature vector and pattern parameter table of the optimal garment design scheme. Then, it converts the Pantone color codes of the fabrics in the fabric database, such as 18-3940 TCX, into 3D RGB vectors, and uses One-Hot encoding for texture types to create 4D vectors. These are then concatenated into a visual attribute vector. A cosine similarity algorithm is used to calculate the similarity value between the style feature vector and the visual attribute vector, and fabrics with a similarity value ≥ 0.7 are selected, such as "Pantone color code 18-3940 TCX". The fabrics with "jacquard texture" were selected; then, physical property matching was performed: according to the preset adaptation rules of the pattern type in the pattern parameter table (elastic modulus ≥ 2.0 cN·dtex⁻¹, air permeability ≥ 120 mm / s, weight 150-200 g / m²), the physical properties of the selected fabrics were compared, and it was determined that the fabric with "elastic modulus 2.2 cN·dtex⁻¹, air permeability 130 mm / s, weight 180 g / m²" met the adaptation requirements; finally, the fabric was comprehensively analyzed. Matching score calculation: The overall fabric matching score is calculated based on a similarity score of 40% and a suitability score of 60%. The overall fabric matching score of this fabric is 0.92. This fabric with the highest overall matching score is selected as the candidate fabric. The fabric information, including "Fabric Identifier: MAT-24-000012; Physical Properties: Elastic Modulus 2.2cN·dtex⁻¹, Air Permeability 130mm / s, Weight 180g / m²; Visual Properties: Pantone Color Code 18-3940 TCX, Jacquard Texture", is output and transmitted to the virtual try-on module.

[0025] The virtual try-on module receives the optimal clothing design scheme (3D clothing model, pattern parameter table) from the intelligent design generation module and candidate fabric information from the fabric matching module. First, it uses the ICP algorithm to align the coordinates of the OBJ format 3D clothing model with the 3D human body model, controlling the alignment error to ≤0.3mm. Then, based on the pattern parameter table, it assigns differentiated weights to the mesh vertices of the shoulders, waist, hips, and sleeves using mesh vertex weight adjustment technology, achieving a static fit between the 3D clothing model and the 3D human body model. Next, it calls the PBD physics simulation engine, inputting the elastic modulus and weight of the candidate fabrics, to simulate the dynamic wrinkling deformation of the clothing under the user's posture of raising their arm 30° and turning 45° (the wrinkle depth at the armhole is ≤2cm when raising the arm, and the skirt's drape curvature conforms to the law of gravity when turning). Finally, it uses PBR technology to match the Pantone color number 18-3940 of the candidate fabrics. The TCX data is converted to RGB color parameters (R=245, G=240, B=230), the jacquard texture is loaded as a 2048×2048px resolution texture map, and the 3D clothing model is rendered using an ambient occlusion algorithm to generate a 360-degree rotatable 3D virtual try-on effect, which is then transmitted to the effect display unit of the interactive module.

[0026] When the solution optimization module is working, the effect display unit of the interaction module displays the virtual try-on effect of the clothing in a 360-degree rotatable 3D screen. Users use the graphical annotation tool of the feedback collection unit to annotate feedback such as "sleeve length is too short, needs to be lengthened by 3cm". After receiving this feedback, the solution optimization module calls the YOLOv8 object detection algorithm to identify "sleeve length (part to be modified)" and extract "lengthen by 3cm (adjustment parameter)". Then, it maps the part to be modified and the adjustment parameter to the "sleeve length" field of the structured requirements document, generating an updated structured requirements document (sleeve length adjusted from 50cm to 53cm). Then, based on the updated structured requirements document, it regenerates the parameter vector and combines it with the original... The human feature vectors of the 3D human model are fused into a new fusion vector, which is then input into the VAE generation model to regenerate three candidate clothing design schemes. The new optimal clothing design scheme with "sleeve length 53cm and other parameters unchanged" is selected according to the original matching score rules. Then, the scheme optimization module calls the fabric matching module to confirm the candidate fabrics (elastic modulus and breathability meet the pattern requirements after the sleeve length is increased), and generates a new 3D clothing virtual try-on effect through the virtual try-on module. Finally, the interaction module displays the new try-on effect. If the user confirms that there is no need for modification, the scheme optimization module stops iterating and outputs the final optimal clothing design scheme, candidate fabric information, and virtual try-on effect, completing this personalized clothing design.

[0027] The above description describes specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A data model-based intelligent interactive clothing design system, characterized in that, include: The interactive module is used to acquire users' human body data and clothing design requirements, display the virtual try-on effect of clothing, and receive user feedback for modification. The system comprises four modules: a human body modeling module to generate 3D human body models based on human body data and extract human body size parameters and posture feature labels; a demand intelligent parsing module to transform clothing design requirements into a structured requirement document containing style tags and pattern constraint parameters, based on posture feature labels; an intelligent design generation module to input the 3D human body model and structured requirement document into a VAE generation model to generate the optimal clothing design scheme; a fabric matching module to build a fabric database and match corresponding candidate fabrics in the database according to the optimal clothing design scheme, outputting the fabric information of the corresponding candidate fabrics; a virtual try-on module to receive the optimal clothing design scheme and fabric information, realize dynamic fitting simulation between the 3D clothing model and the 3D human body model, and generate a virtual try-on effect; and a scheme optimization module to optimize the optimal clothing design scheme based on user feedback until it meets user needs.

2. The intelligent interactive clothing design system based on a data model according to claim 1, characterized in that, The interactive module includes a data receiving unit, an effect display unit, and a feedback collection unit. The data receiving unit acquires the user's human body data through a 3D scanning device and obtains the user's clothing design requirements described in natural language through a text input box. The human body data is a PLY format point cloud file. The effect display unit is used to display the virtual try-on effect of clothing in a 360-degree rotatable 3D image. The feedback collection unit provides a graphical annotation tool for users to annotate the parts that need to be modified in the 3D image of the virtual try-on effect and input the adjustment parameters of the parts to be modified in order to collect user feedback.

3. The intelligent interactive clothing design system based on a data model according to claim 2, characterized in that, The system includes a point cloud preprocessing unit, a 3D reconstruction unit, and a feature extraction unit. The point cloud preprocessing unit sequentially performs noise point removal, coordinate registration, and point cloud smoothing on the human body data to obtain standardized human body point cloud data. The 3D reconstruction unit constructs a closed triangular mesh model based on the standardized human body point cloud data using the Poisson surface reconstruction algorithm, and then optimizes the details of the closed triangular mesh model by combining human anatomical features to generate a 3D human body model. The feature extraction unit is used to extract human body size parameters and body shape feature labels from the 3D human body model. The human body size parameters include at least chest circumference, waist circumference, hip circumference, shoulder width, back length, sleeve length, and upper thigh length. The body shape feature labels include body type labels, shoulder shape labels, and waist feature labels.

4. The intelligent interactive clothing design system based on a data model according to claim 3, characterized in that, The method for obtaining the structured requirement document is as follows: Entity recognition is performed on the natural language clothing design requirements using the BERT-base-chinese model to obtain a set of requirement keywords containing style keywords and pattern keywords. A pre-built rule engine is then called to process the associated requirement keyword set and body feature tags to generate a structured requirement document containing style tags and pattern constraint parameters. The rule engine modifies and constrains the associated requirement keyword set and body feature tags based on the matching logic in a pre-built clothing knowledge graph. The matching logic of the clothing knowledge graph is a rule for adjusting the set value of the pattern parameter when the body feature tag corresponds to the pattern keyword.

5. The intelligent interactive clothing design system based on a data model according to claim 4, characterized in that, The process of obtaining the optimal clothing design scheme is as follows: the 3D human body model is converted into a feature vector containing human body size parameters, and the structured requirements document is converted into a parameter vector. The feature vector and parameter vector are fused into a fused vector and then input into the VAE generation model to generate multiple candidate clothing design schemes. The cosine similarity between the style feature vector of each candidate clothing design scheme and the style tag in the structured requirements document is calculated. Combined with the fit between the pattern parameter table and the pattern constraint parameters, the candidate clothing design scheme with the highest matching score is selected as the optimal clothing design scheme. Each candidate clothing design scheme includes a pattern parameter table, a 3D clothing model and a style feature vector.

6. The intelligent interactive clothing design system based on a data model according to claim 5, characterized in that, The fabric database stores the physical and visual attributes of each fabric, and each fabric corresponds to a unique fabric identifier. The candidate fabrics are selected by extracting style feature vectors and pattern parameter tables from the optimal clothing design scheme, matching the style feature vectors with visual attributes for similarity, and matching the pattern types in the pattern parameter table with physical attributes for adaptability. Based on the results of similarity matching and adaptability matching, a comprehensive fabric matching score is calculated, and the fabric with the highest comprehensive fabric matching score is selected. The fabric information includes the fabric identifier, physical attributes, and visual attributes of the candidate fabrics.

7. The intelligent interactive clothing design system based on a data model according to claim 6, characterized in that, The virtual try-on module uses the ICP algorithm to align the coordinates of the 3D clothing model and the 3D human body model of the optimal clothing design scheme. Based on the pattern parameter table of the optimal clothing design scheme, it uses mesh vertex weight adjustment technology to achieve static fitting between the 3D clothing model and the 3D human body model. Accordingly, it calls the PBD physical simulation engine to perform dynamic fitting simulation in combination with the physical properties of the candidate fabrics. It also uses PBR technology to convert the visual properties of the candidate fabrics into rendering parameters of the 3D clothing model, generating a three-dimensional virtual try-on effect.

8. The intelligent interactive clothing design system based on a data model according to claim 7, characterized in that, The process of optimizing the clothing design scheme is as follows: the user-marked parts that need to be modified are identified by the object detection algorithm, and the adjustment parameters in the user's modification feedback of the parts that need to be modified are extracted. The parts that need to be modified and the adjustment parameters are mapped to the corresponding fields in the structured requirements document to generate an updated structured requirements document. Based on the updated structured requirements document, the parameter vector is regenerated and fused with the feature vector of the original 3D human body model to form a new fusion vector. This new fusion vector is then input into the VAE generation model to regenerate candidate clothing design schemes and select the optimal clothing design scheme.

9. A data model-based intelligent interactive clothing design method, the method being used to implement the data model-based intelligent interactive clothing design system of claim 1, the method comprising the following steps: S1. Obtain the user's anatomy data and clothing design requirements; S2. Generate a 3D human body model based on human body data, and extract the human body size parameters and body shape feature labels from the 3D human body model; S3. Combine body shape feature tags to transform clothing design requirements into a structured requirements document that includes style tags and pattern constraint parameters; S4. Input the 3D human body model and structured requirements document into the VAE to generate the optimal clothing design scheme. Based on this, match the candidate fabrics corresponding to the optimal clothing design scheme from the fabric database and output the fabric information. Then, based on the optimal clothing design scheme and fabric information, realize the dynamic fitting simulation between the 3D clothing model and the 3D human body model, generate and display the virtual try-on effect of the clothing. S5. Optimize the best clothing design based on user feedback regarding the virtual try-on effect until the user's needs are met.

Images