Ai-based virtual garment fitting system
The AI-based virtual clothing fitting system addresses inaccuracies in conventional systems by generating a 3D avatar and simulating clothing interaction with precise deformation and pressure analysis, enhancing the fitting experience and reducing returns.
Patent Information
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- LACE DESIGN CO CO LTD
- Filing Date
- 2025-08-25
- Publication Date
- 2026-07-21
AI Technical Summary
Conventional virtual fitting systems struggle to accurately simulate how clothing interacts with a user's body, failing to dynamically reproduce garment deformation and pressure points due to reliance on simple modeling or static rule-based methods, leading to discrepancies with actual fitting and reduced visual persuasiveness.
An AI-based virtual clothing fitting system that generates a 3D avatar from user body measurements, fits 3D clothing to the avatar, and quantitatively analyzes deformation and pressure through a simulation unit using a physics-based engine, incorporating AI prediction for future body changes.
Provides accurate virtual fitting results similar to actual wear, reducing return rates and increasing consumer satisfaction by allowing precise simulation of garment deformation, curvature, and contact pressure, and predicting future body shape changes.
Smart Images
Figure 112025096670481-PAT00001_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to an artificial intelligence-based virtual clothing fitting system. Background Technology
[0002] Generally, in online shopping environments, consumers make purchasing decisions based on product images or size information due to the limitation that they cannot actually try on clothing products. However, since it is difficult to accurately judge comfort, fit, and material characteristics based solely on information available on the screen, problems such as increased return rates, decreased consumer satisfaction, and rising operating costs for sellers frequently occur.
[0003] To address these issues, virtual fitting technology, which virtually simulates how a user would look wearing clothing based on their body measurements, is gaining attention. This technology involves creating a 3D avatar, applying digital clothing to it, and then visually rendering the silhouette and movement of the garment through simulation. Through this, consumers can preview the fit and style of the clothing before actually wearing it, and improve the accuracy of their size selection.
[0004] Conventional virtual fitting systems have generally relied on simple modeling-based 3D rendering or predefined template matching methods. In this case, it is difficult to accurately reflect the user's body information, and there are limitations in precisely implementing actual wearing effects, such as the physical properties, deformation, and contact pressure of the clothing. Furthermore, when the process of simulating the interaction between the avatar and the clothing is performed using numerical calculations or static rule-based methods, there are problems such as large discrepancies with the actual fitting situation and reduced visual persuasiveness.
[0005] In particular, factors such as the material characteristics, thickness, elasticity, and tailoring structure of a garment significantly influence actual wear, and it is essential to be able to dynamically reproduce how the garment deforms and where pressure is applied depending on the user's body shape or posture. However, existing technologies often fail to precisely reflect these dynamic elements and frequently substitute them with simple shape overlays or animation processing, making it difficult to provide accurate fitting results tailored to the user's body shape.
[0006] Therefore, there is a growing need for a high-precision virtual clothing fitting system that can automatically extract precise body information based on a user's body image, generate a 3D avatar corresponding to the actual body shape based on this information, and fit 3D clothing composed of various clothing blueprints to the avatar in real time to dynamically simulate quantitative information such as the elongation rate, curvature, and contact pressure of the clothing. The problem to be solved
[0007] The objective of the present invention is to provide an AI-based virtual clothing fitting system that solves the problems associated with the aforementioned prior art. By generating a virtual avatar based on a user's body information, digitizing various garments, and virtually fitting them onto the avatar, the system visually implements fitting results similar to actual wear and is configured to quantitatively analyze wearing responses, such as garment deformation patterns and body-fit characteristics. This allows consumers to experience trying on clothing that fits their body shape in an online environment, reduces uncertainty in size selection, lowers return rates, and improves user satisfaction. means of solving the problem
[0008] The above objective is achieved, according to the present invention, by a body information generating unit that generates body information including the height, chest circumference, waist circumference, and hip circumference of a user based on a body image of a user captured by a 3D camera; an avatar generating unit that generates a 3D avatar based on the body information; a garment generating unit that generates a 3D garment based on a previously stored garment drawing; and a simulation unit that fits the 3D garment to the 3D avatar to generate simulation information indicating the elongation rate, curvature, and contact pressure of the 3D garment. The system includes a rendering unit that visually outputs the 3D avatar fitted with the 3D clothing based on the simulation information, and the body information generation unit includes a preprocessing module that generates 3D point cloud data from the body image and preprocesses the body image by performing background removal, noise filtering, surface refinement, and alignment processing on the point cloud data; an extraction module that inputs the preprocessed body image into an artificial intelligence model based on a PointNet architecture to extract feature points from the point cloud data, wherein if the prediction reliability of the feature points falls below a preset threshold, the feature points are discarded or classified as targets for re-estimation; and a computation module that generates the body information by performing a consistency check by comparing the distance between the feature points with standard human body ratios, determining feature points exceeding a preset tolerance as errors, and calculating the user's height, chest circumference, waist circumference, and hip circumference based on the distance between feature points that passed the consistency check. The simulation unit calculates the elongation rate, the curvature, and the contact pressure by placing the 3D clothing on the surface of the 3D avatar, wherein the elongation rate is the 3D Calculated based on the degree of surface deformation of the garment, the curvature is calculated based on the change in curved surface inclination and folding distribution of the 3D garment, and the contact pressure is calculated based on the contact position and change in contact distance between the 3D garment and the 3D avatar surface.The simulation unit generates predictive simulation information reflecting future body shape changes of the 3D avatar through an artificial intelligence prediction module comprising a recurrent neural network (RNN) with a Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU) structure that performs time-series-based body shape change modeling based on user information including health status information, age information, weight change trend information, pregnancy status information, gestational week information, hormone change history information, and seasonal environmental information; the artificial intelligence prediction module generates a 3D avatar at a future point in time by reflecting body shape changes of the user's abdomen, hips, and thighs according to the gestational week information when the user is pregnant; calculates the elongation rate, curvature, and contact pressure of the 3D clothing for the 3D avatar at the future point in time to generate the predictive simulation information; generates the 3D avatar at the future point in time by reflecting body shape changes of the abdomen and upper body due to reduced outdoor activity and weight gain in winter based on the seasonal environmental information, and reflecting body shape reduction due to increased sweat elimination and heat dissipation efficiency in summer; and the 3D at the future point in time This is achieved by an AI-based virtual clothing fitting system that calculates the elongation rate, curvature, and contact pressure of the 3D clothing for the avatar to generate the prediction simulation information, and the rendering unit generates a rendering frame at a current time point based on the simulation information and generates a rendering frame at a future time point based on the prediction simulation information, and simultaneously outputs the rendering frame at the current time point and the rendering frame at the future time point, while applying enlarged display or color contrast to mesh areas where the difference in elongation rate, curvature, and contact pressure between time points exceeds a preset reference value.
[0009] In addition, the avatar generation unit is characterized by generating a 3D avatar whose shape can be adjusted according to the body information through a parametric human body model.
[0010] delete
[0011] In addition, the garment generation unit is characterized by generating 3D panels for each part of the garment based on the garment drawing, and combining a plurality of the 3D panels according to the seam lines of the garment indicated on the garment drawing to generate the 3D garment.
[0012] delete Effects of the invention
[0013] According to the present invention, digital clothing can be virtually fitted to a three-dimensional avatar generated based on a user's body information, and the results can be visually provided while simultaneously quantitatively analyzing fitting responses such as deformation patterns or tightness characteristics of the clothing. This allows for a fitting experience similar to actual wearing to be provided in an online environment, and consumers can more accurately select clothing suitable for their body shape, thereby reducing return rates and increasing purchase satisfaction.
[0014] In addition, various functions such as size recommendation, fit prediction, and wear simulation can be automated based on the results of quantitative analysis of wear response. This can be utilized in various fields such as clothing design, virtual fitting shopping malls, and personalized recommendation systems, thereby contributing to the improvement of user experience as well as distribution efficiency and the digital transformation of the entire industry.
[0015] Meanwhile, the effects of the present invention are not limited to those mentioned above, and various effects may be included within the scope obvious to a person skilled in the art from the contents described below. Brief explanation of the drawing
[0016] FIG. 1 illustrates the connections between the components of an artificial intelligence-based virtual clothing fitting system according to an embodiment of the present invention, and FIG. 2 illustrates the connections between the components of a body information generation unit of an artificial intelligence-based virtual clothing fitting system according to an embodiment of the present invention, and FIG. 3 illustrates an example of three-dimensional point cloud data of a preprocessing module of a body information generation unit of an artificial intelligence-based virtual clothing fitting system according to an embodiment of the present invention. Specific details for implementing the invention
[0017] Hereinafter, some embodiments of the present invention will be described in detail with reference to the exemplary drawings. It should be noted that in assigning reference numerals to the components of each drawing, the same components are given the same reference numeral whenever possible, even if they are shown in different drawings.
[0018] In addition, when describing embodiments of the present invention, if it is determined that a detailed description of related known configurations or functions would hinder understanding of the embodiments of the present invention, such detailed description is omitted.
[0019] In addition, terms such as first, second, A, B, (a), (b), etc., may be used when describing the components of the embodiments of the present invention. These terms are used merely to distinguish the components from other components, and the essence, order, or sequence of the components is not limited by the terms used.
[0020] Additionally, in the specification, the singular form includes the plural form unless specifically stated otherwise in the text. The terms “comprising” and / or “comprising” as used in the specification do not exclude the presence or addition of one or more other components in addition to the mentioned components.
[0021] Additionally, in describing embodiments of the present invention, each "part," "module," or "step" may be implemented through a processor and memory. The term "processor" should be broadly interpreted to include general-purpose processors, central processing units (CPUs), microprocessors, digital signal processors (DSPs), controllers, microcontrollers, state machines, and the like. In some contexts, the term "processor" may refer to an application-specific integrated circuit (ASIC), a programmable logic device (PLD), a field programmable gate array (FPGA), and the like. The term "processor" may also refer to a combination of processing devices, such as, for example, a combination of a DSP and a microprocessor, a combination of multiple microprocessors, a combination of one or more microprocessors combined with a DSP core, or any other combination of such configurations.
[0022] Furthermore, memory should be broadly interpreted to include any electronic component capable of storing electronic information. Memory may also refer to various types of processor-readable media such as random access memory (RAM), read-only memory (ROM), non-volatile random access memory (NVRAM), programmable read-only memory (PROM), eraseable-programmable read-only memory (EPROM), electrically eraseable PROM (EEPROM), flash memory, magnetic or optical information storage devices, registers, etc. If a processor can read information from memory or write information to memory, memory is said to be in an electronic communication state with the processor, and each "part," "module," or "stage" may be implemented through a program or application based on the processor and memory in an electronic communication state.
[0023] Furthermore, in the present invention, Artificial Intelligence (AI) refers to a technology that imitates human learning ability, reasoning ability, and perceptual ability, and implements them on a computer, and may include concepts such as machine learning and symbolic logic. Machine Learning (ML) is an algorithmic technology that classifies or learns the characteristics of input information on its own. AI technology can analyze input information as a machine learning algorithm, learn from the results of that analysis, and make judgments or predictions based on the results of that learning. Additionally, technologies that mimic the functions of the human brain, such as cognition and judgment, by utilizing machine learning algorithms can also be understood as falling within the category of AI. For example, technological fields such as linguistic understanding, visual understanding, reasoning / prediction, knowledge representation, and motion control may be included.
[0024] Artificial intelligence learning models or neural network models can be designed to implement the structure of the human brain on a computer and may include multiple network nodes that simulate neurons of a human neural network and have weights. These multiple network nodes can have interconnected relationships by simulating the synaptic activity of neurons, where neurons exchange signals through synapses. In an artificial intelligence learning model, multiple network nodes can be located in layers of different depths and exchange information according to convolutional connections. An artificial intelligence learning model may be, for example, an Artificial Neural Network (ANN) or a Convolutional Neural Network (CNN). An artificial intelligence learning model can be machine learned according to methods such as supervised learning, unsupervised learning, and reinforcement learning. Machine learning algorithms that can be used to perform machine learning include Decision Trees, Bayesian Networks, Support Vector Machines, Perceptrons, and Clustering.
[0026] From now on, an artificial intelligence-based virtual clothing fitting system (100) according to one embodiment of the present invention will be described in detail with reference to the attached drawings.
[0027] FIG. 1 illustrates the connections between components of an artificial intelligence-based virtual clothing fitting system according to an embodiment of the present invention, FIG. 2 illustrates the connections between components of a body information generation unit of an artificial intelligence-based virtual clothing fitting system according to an embodiment of the present invention, and FIG. 3 illustrates an example of 3D point cloud data of a preprocessing module of a body information generation unit of an artificial intelligence-based virtual clothing fitting system according to an embodiment of the present invention.
[0028] As illustrated in FIG. 1, an artificial intelligence-based virtual clothing fitting system (100) according to one embodiment of the present invention includes a body information generation unit (110), an avatar generation unit (120), a clothing generation unit (130), a simulation unit (140) and / or a rendering unit (150).
[0029] The body information generation unit (110) generates body information including the user's height, chest circumference, waist circumference, and hip circumference based on a body image of the user captured by a 3D camera, and is electrically connected to the avatar generation unit (120).
[0030] As illustrated in FIG. 2, the body information generation unit (110) includes a preprocessing module (111), an extraction module (112), and / or a calculation module (113).
[0031] The preprocessing module (111) preprocesses images of the user's body captured by a 3D camera and is electrically connected to the extraction module (112).
[0032] Specifically, the preprocessing module (111) can provide a data environment that can extract the body shape with high precision by generating three-dimensional point cloud data representing the external shape information of the body from a user’s body image captured by a 3D camera, and performing processing such as background removal, noise filtering, surface refinement, and alignment from the point cloud.
[0033] As illustrated in FIG. 3, 3D point cloud data is constructed by converting a user's body image into multiple point data in 3D space, and each point may include coordinate values (x, y, z) representing a specific location on the surface of the user's body.
[0034] The preprocessing module (111) can receive a point cloud as input data, select only valid points that accurately reflect the body shape, and perform a series of processes to adjust density and alignment.
[0035] The preprocessing module (111) can apply distance-based filtering, color segmentation, and / or dense outlier removal techniques to remove background elements and image noise present on the point cloud.
[0036] For example, points that fall outside the standard distance range from the body surface or have a significantly low spatial density with surrounding points may be identified as noise caused by external objects or measurement errors and removed.
[0037] Afterwards, the preprocessing module (111) can perform surface correction work based on the refined point cloud to ensure continuity and consistency of the body surface shape.
[0038] Surface correction operations may include moving average-based smoothing (Moving Least Squares Smoothing), mesh reconstruction, and / or hole filling, which can minimize distortion of body curves and restore seamlessly connected shape information.
[0039] In addition, the preprocessing module (111) can perform alignment processing, such as estimating the center axis of the body, aligning the reference plane, and / or correcting rotation, to eliminate coordinate system deviations that may occur depending on the user's shooting posture or direction.
[0040] Alignment processing ensures left-right symmetry based on the front of the body shown in the body image and normalizes the body data according to the reference direction, thereby enabling consistent analysis and comparison of feature points (P) in subsequent steps.
[0041] The extraction module (112) extracts feature points (P) of the user's body based on a preprocessed body image and is electrically connected to the preprocessing module (111) and / or the computation module (113).
[0042] Specifically, the extraction module (112) receives three-dimensional point cloud data transmitted from the preprocessing module (111) as input and can automatically detect feature points (P) representing reference positions of the user's body.
[0043] The feature points (P) may consist of points located at major points constituting the user's body, such as the crown of the head, the center of the neck, the left shoulder, the right shoulder, the left elbow, the right elbow, the left wrist, the right wrist, the center of the chest, the center of the waist, the left hip, the right hip, the left knee, the right knee, the left ankle and / or the right ankle, and each feature point (P) may be represented as a single point containing x, y, and z coordinate values.
[0044] The extraction module (112) may include an artificial intelligence model based on the PointNet architecture and can predict the location of each feature point (P) in the input point cloud.
[0045] PointNet can be designed with a structure capable of processing unstructured point cloud data as input, and can simultaneously learn the local characteristics of specific feature points (P) and the global structure of the entire point cloud.
[0046] Each feature point (P) can be mapped to a high-dimensional feature space by a Multi-Layer Perceptron (MLP), and then the MLP can be applied in parallel to all points.
[0047] Max pooling, a symmetric function, is performed on the generated point-specific feature vectors to generate a fixed-length global feature vector representing the entire point cloud.
[0048] Max pooling can be performed by extracting the maximum value from the entire point cloud for each feature dimension and is not affected by the order of the input points. That is, since the same global feature vector is generated even if the order of the points is changed, permutation invariance can be guaranteed. This characteristic is suitable for 3D point cloud data that is not ordered and allows feature points (P) to be extracted according to the same criteria even with different postures for each user or changes in the user's posture.
[0049] The global feature vector is passed to a subsequent regression circuit, and coordinate values of each feature point (P) are output according to the learned weights. The output values may include the x, y, and z coordinates of each feature point along with a prediction confidence score, and if the confidence score falls below a preset threshold, the result may be discarded or classified for re-estimation.
[0050] A PointNet-based artificial intelligence neural network can be pre-trained with a training point cloud dataset containing 3D body scan data.
[0051] In the training data, accurate coordinate values for each feature point are assigned as labels, and the model can be trained by applying the Mean Squared Error (MSE) as the loss function. The trained model has fixed parameters and can extract feature points in real time during the inference process described later.
[0052] In the inference process, a preprocessed point cloud is received as input, and the coordinates of each feature point (P) and the prediction confidence of each feature point can be output.
[0053] The extracted feature points (P) undergo an internal consistency check, and the consistency check can be performed based on distance and direction information between the feature points (P). For example, key reference distances, such as the horizontal distance between the left and right shoulders, the vertical distance between the center of the chest and the center of the waist, and the length between the knee and the ankle, can be calculated and compared with predefined standard human body proportions.
[0054] At this time, feature points (P) exceeding the tolerance are considered errors and may be invalidated or passed to a correction routine, and the consistency evaluation can be performed through Euclidean distance calculation, vector direction analysis, and average distance difference evaluation.
[0055] Afterwards, the final output data is organized into a structured form and may include information on the name of each feature point (P), 3D coordinate values (x, y, z), and prediction reliability, and the final output data is transmitted to the calculation module (113) described later and used to calculate height, chest circumference, waist circumference, and hip circumference.
[0056] The PointNet-based structure of the extraction module (112) can effectively process unordered point clouds and is designed to reflect both global characteristics and local structures of the body shape, so it can serve as a key technical element that improves the accuracy of avatar generation and the reliability of clothing fitting results.
[0057] The calculation module (113) generates body information by calculating the user's height, chest circumference, waist circumference, and hip circumference based on the distance between feature points (P) extracted from the extraction module (112), and is electrically connected to the extraction module (112).
[0058] The operation module (113) can perform procedures such as distance calculation, cross-sectional analysis, and / or curve approximation based on the relative positional relationship of a specific plurality of feature points (P).
[0059] Height can be calculated based on the coordinate values of the crown feature point (P), the left ankle feature point, and the right ankle feature point (P).
[0060] Height can be calculated by averaging the vertical (y-axis) coordinates among the coordinate values of the left and right ankle feature points (P) to set a lower reference point, and calculating the absolute value of the difference between the reference point and the y-axis coordinate value of the crown of the head feature point (P).
[0061] The calculation method at this time can be performed according to the Euclidean distance calculation formula based on straight-line distance, and even if the body is tilted due to the user's posture, vertical reference alignment can be applied to derive a value similar to the actual height.
[0062] Chest circumference can be calculated based on the coordinate values of the left shoulder feature point, the right shoulder feature point, and / or the center of the chest feature point.
[0063] The length of the major axis of the horizontal cross-section of the chest region is set based on the distance between the coordinate values of the left and right shoulder feature points, and the direction of the minor axis can be estimated by analyzing the direction and length of two vectors pointing from the chest center feature point toward the left and right shoulders.
[0064] In this case, the horizontal cross-section of the chest area can be approximated as an ellipse, and the chest circumference can be calculated by applying the approximate ellipse circumference formula based on the lengths of the major and minor axes.
[0065] Waist circumference can be calculated based on the coordinate values of the waist center feature point.
[0066] A horizontal cross-section of the waist area of the body is set based on the coordinate values of the waist center feature point, but when both the left waist feature point and the right waist feature point are provided, the distance between the coordinate values of the left waist feature point and the coordinate values of the right waist feature point can be calculated to set the length of the major axis of the horizontal cross-section of the waist area.
[0067] The horizontal cross-section of the waist area can be approximated as an ellipse, and the waist circumference can be calculated by applying the approximate elliptical circumference formula based on the lengths of the major and minor axes. The major axis is set as the distance between the coordinates of the left waist feature point and the right waist feature point, and the minor axis can be calculated based on the standard curvature distribution defined in the anterior-posterior direction from the coordinates of the waist center feature point.
[0068] If left and right waist feature points are not provided and only the waist center feature point is provided, the waist circumference can be calculated according to the same elliptical approximation method by referring to a predefined body type classification result based on the coordinate values of the waist center feature point, applying an average section coefficient corresponding to the classified body type, deriving correction values for the major and minor axes.
[0069] Hip circumference can be calculated based on the coordinate values of the left hip feature point and the right hip feature point.
[0070] The length of the major axis of the horizontal cross-section of the buttocks can be set by calculating the distance between the coordinate values of the left buttock feature point and the right buttock feature point, and the direction of the minor axis can be estimated by analyzing the vector distribution in the direction of the left and right buttock feature points based on the coordinate values of the buttock center feature point.
[0071] The horizontal cross-section of the hip area can be approximated as an ellipse, and the hip circumference can be calculated by applying the approximate ellipse circumference formula based on the lengths of the major and minor axes.
[0072] The major axis can be defined as the distance between the coordinate values of the left hip feature point and the right hip feature point, and the minor axis can be set based on the reference curvature distribution in the anterior-posterior direction from the coordinate values of the hip center feature point.
[0073] In addition, if there is a tilt or rotation between the hip feature points depending on the user's posture, the coordinate system can be normalized based on the coordinate values of the hip center feature point, and after correcting the major and minor axis lengths on the reference horizontal plane, the hip circumference can be calculated according to the elliptical approximation method.
[0074] The calculation module (113) can generate body information based on the calculated height, chest circumference, waist circumference, and hip circumference, and transmit it to the avatar generation unit (120) to be described later.
[0075] In addition, the calculation module (113) can quantitatively evaluate the distance and direction between feature point coordinate values and perform a consistency evaluation to verify the consistency of the calculated result.
[0076] The inspection items for the conformity evaluation may include left-right symmetry, vertical alignment, and the ratio of distances between parts, and an error flag may be set if the allowable error range is exceeded when compared to a predefined standard human body ratio. If an error is detected, correction for the corresponding dimension item is performed based on the average value of the nearest similar body shape data, and in the case of a fatal error, a signal for re-extracting the feature point (P) may be transmitted to the extraction module (112).
[0077] The calculation module (113) can accurately calculate body measurements in a consistent and automated manner without receiving direct input from the user through calculations based on the coordinate values of feature points. This improves the level of automation and precision of the entire system, and has the effect of substantially improving the accuracy of avatar modeling and the reliability of clothing fitting results.
[0078] The avatar generation unit (120) generates a 3D avatar based on body information received from the body information generation unit (110), and is electrically connected to the body information generation unit (110), the simulation unit (140), and / or the rendering unit (150).
[0079] Specifically, the avatar generation unit (120) receives user body information including height, chest circumference, waist circumference, and hip circumference as input and can automatically generate a three-dimensional human body avatar model similar to the user's body shape.
[0080] The avatar generation unit (120) can generate a 3D avatar whose shape can be adjusted according to body information based on a parametric human body model designed to accommodate various body types.
[0081] A parametric human body model is a three-dimensional human body model designed to quantitatively generate and control various body types by representing the human body with mathematical parameters.
[0082] A parametric human body model can automatically adjust the overall human body shape by receiving key measurements such as height, chest circumference, waist circumference, and hip circumference as input values and reflecting them in control parameters that constitute the shape of each body part.
[0083] In addition, since parametric human body models can efficiently implement various body shape deformations on a base mesh of the same structure, they can be effectively utilized to create virtual avatars that precisely reflect the actual user's body shape.
[0084] For example, height affects spine and leg length parameters, chest circumference adjusts the radius, curvature, and protrusion values of the outer curve of the chest, and waist and hip circumference can be reflected in the cross-sectional deformation modulus forming the shape of the abdomen and buttocks.
[0085] The avatar generation unit (120) can numerically calculate shape parameters corresponding to each part of the body and transmit them to a modeling engine to generate a 3D avatar in the form of a 3D mesh. At this time, the movement of vertices of the mesh, polygon expansion, and surface curvature adjustment can be performed automatically, and can be implemented to match the body information containing information about the user's full body proportions and body shape.
[0086] Additionally, the avatar generation unit (120) may refer to a standard human body dataset to improve the visual realism and body similarity of the model. The standard dataset includes 3D human body data for various ages, genders, and body types, and the generation accuracy can be improved by setting an initial avatar shape based on a standard body type most similar to the input body information, and then configuring a correction model by reflecting the user's body information on it.
[0087] In addition, the avatar generation unit (120) can perform a conformity check on the shape of the generated 3D avatar.
[0088] It is determined whether the dimension items of the 3D avatar match the body information received from the body information generation unit (110) within a certain error range, and if the error exceeds the allowable standard, the model can be automatically reconstructed by repeatedly applying a correction routine. This alignment check can be performed in multiple stages not only for dimension comparison but also for cross-sectional shape, ratio, continuity, etc.
[0089] The 3D avatar generated in the avatar generation unit (120) can be defined as model data consisting of a surface mesh, vertex information, skeleton structure and / or body shape parameters, and can be converted into a standard 3D file format and used for virtual clothing fitting simulation and visual output through the simulation unit (140) and rendering unit (150).
[0090] In addition, the generated 3D avatar may include structured internal metadata such as body shape code, applied parameters, number of vertices, and cross-sectional coordinate system information, which can support efficient computation and physics-based simulation.
[0091] The avatar generation unit (120) can perform a core function that can precisely simulate the user's actual wearing condition during the virtual fitting process by automatically generating a highly accurate 3D human body model based on the individual user's body shape information, thereby substantially improving the precision and persuasiveness of the body shape-tailored clothing simulation.
[0092] The garment creation unit (130) creates a 3D garment based on a previously stored garment drawing and is electrically connected to the simulation unit (140) and / or rendering unit (150).
[0093] The garment creation unit (130) stores a garment drawing, which is a production drawing of a real garment expressed in a 2D plane, and the garment drawing may include pattern information of multiple garment parts that constitute the garment.
[0094] Pattern information for each part of the garment may include cutting line coordinate information, contour curvature information, sewing line information, sewing sequence information and / or dimension information, and may be provided in the form of DXF (Drawing Exchange Format), SVG (Scalable Vector Graphics), or a CAD-based format for garment design.
[0095] The garment creation unit (130) can create 3D panels for each part of the garment based on a previously stored garment drawing, and create a 3D garment by combining multiple 3D panels according to the seam lines of the garment indicated on the garment drawing.
[0096] The garment generation unit (130) parses pattern information for each part of the garment individually and defines 2D panels based on the outline coordinates of the pattern. Then, each 2D panel can be expanded into 3D space and converted into a 3D panel that takes into account the actual wearing form.
[0097] The garment generation unit (130) can individually parse pattern information of a previously stored garment drawing, define a 2D panel based on the outline, and then expand it into a 3D space to generate a 3D panel.
[0098] The 3D panel is a surface structure in mesh units that constitutes each part of the garment, and may include face information, edge information, and / or vertex information in polygon units.
[0099] The 3D panel may include physical property information to represent the shape of the fabric unit, and the physical property information may include initial physical property values such as thickness, density, tensile strength, and bending stiffness at the time of creation.
[0100] The garment creation unit (130) connects the generated multiple 3D panels according to the joint line information indicated on the garment drawing.
[0101] The joint line information defines the connection boundaries of multiple 3D panels, and joints can be performed using a direct mapping method between vertices.
[0102] During the 3D panel joining process, the difference in distance between boundary vertices of the 3D panels is minimized, the alignment between meshes is maintained, and a structurally stable connection is achieved by considering the joining angle and direction. After the joining process, the entire 3D panel is composed of a single continuous 3D garment.
[0103] The garment generation unit (130) can perform a mesh consistency check on the generated 3D garment.
[0104] Consistency checks are based on criteria such as missing seams, vertex inconsistencies, step differences between meshes, and the presence of non-closed surface structures, and can be automatically analyzed through geometric operations based on vertex coordinates.
[0105] In the event of a problem, the mesh connectivity state can be restored by generating missing joint lines or rearranging vertices through an automatic correction algorithm, and this is performed as an essential procedure to ensure simulation stability.
[0106] The garment generation unit (130) finally sets physical property data for the 3D garment with secured alignment. The physical property data may include data on the material type, surface friction coefficient, elastic modulus, recovery coefficient, thickness and / or density of the garment, and may be individually assigned to each 3D panel constituting the 3D garment.
[0107] The garment generation unit (130) can output the generated 3D garment in a structured mesh format. The output data consists of a mesh identifier, panel configuration information, vertex coordinates, joining structure, hierarchy information between panels, material tag, etc., and the file format can be converted to one of OBJ (Wavefront Object), FBX (FilmBox), or glTF (GL Transmission Format). The output 3D garment is transmitted to the simulation unit (140) to be fitted to a 3D avatar, and then can be visually implemented through the rendering unit (150).
[0108] The garment generation unit (130) accurately generates 3D panels for each garment part based on digital garment drawing data configured according to actual garment design standards, and combines them along seam lines to form a high-precision 3D garment. In addition, by including a matched mesh structure and fabric properties for each panel, it provides a digital garment that can be precisely matched with a 3D avatar, and can increase the precision and reliability of the simulation and visualization overall.
[0109] The simulation unit (140) fits the 3D clothing generated by the clothing generation unit (130) onto the 3D avatar generated by the avatar generation unit (120) to generate simulation information representing the elongation rate, curvature, and contact pressure of the 3D clothing, and is electrically connected to the avatar generation unit (120), the clothing generation unit (130), and the rendering unit (150).
[0110] The simulation unit (140) receives the initial state of the 3D garment and shape data of the 3D avatar as input, and numerically calculates the physical interaction between the two objects at time steps to quantitatively analyze fitting results similar to actual wearing. The simulation is performed through a physics-based simulation engine, and physical equations to which Newton's laws of motion and Hooke's law are applied to the mesh vertex units of the 3D garment are interpreted numerically.
[0111] The simulation unit (140) can generate simulation information including the elongation rate, curvature, and contact pressure of the 3D garment.
[0112] The elongation rate can be calculated based on the degree of surface deformation of the 3D garment.
[0113] The elongation rate is a physical quantity that quantifies the rate of surface deformation that occurs as a 3D garment is worn on a 3D avatar, and can be calculated based on the relative increase rate of the distance between mesh vertices or the area of the 3D garment before and after deformation.
[0114] The simulation unit (140) can calculate the elongation rate by comparing the initial mesh shape of the 3D clothing with the deformed mesh shape after fitting to the 3D avatar and converting the increase in distance between each mesh face or vertex into a percentage.
[0115] For example, if the 3D clothing for the arm area was 200mm before fitting to the 3D avatar, but increased to 240mm after wearing, the elongation rate of that area can be calculated as 20%.
[0116] The elongation rate can be compared to the elastic limit of the fabric to serve as a criterion for determining whether excessive deformation has occurred, and at the same time, it can be visualized using color maps or numerical tags to intuitively convey simulation results.
[0117] Curvature can be calculated based on the change in surface slope and fold distribution of the 3D garment.
[0118] Curvature is a physical quantity that indicates how spatially the surface of a 3D garment is curved, the change in surface slope refers to the angular difference between the normal vectors of adjacent mesh faces, and the fold distribution indicates how concentrated high-curvature regions exist.
[0119] The simulation unit (140) can analyze the surface information of a local patch composed of adjacent vertices for a mesh vertex of a 3D garment and calculate the curvature of each vertex using a mathematical model such as mean curvature or Gaussian curvature.
[0120] For example, when a 3D avatar takes a sitting pose, the 3D clothing at the knee area folds rapidly and the curvature value at the mesh vertices of that area spikes to 1.8 / cm or more, which indicates that folding has occurred and can be visually displayed as a curve expression or wrinkle emphasis.
[0121] Curvature can be utilized for predicting wrinkle occurrence locations, optimizing cutting, and verifying quality, and areas exceeding standard values can be automatically marked as regions requiring design changes.
[0122] Contact pressure can be calculated based on the contact location and changes in contact distance between the 3D clothing and the 3D avatar surface.
[0123] Contact pressure is a value normalized by a unit area of the intensity of the repulsive force generated at the point where two surfaces come into contact when a 3D garment is fitted to a 3D avatar, and physically, it can be calculated based on changes in contact stiffness and penetration depth.
[0124] The simulation unit (140) can track the relative position between the mesh vertices of the 3D clothing and the surface vertices of the 3D avatar in time units and numerically calculate the change in contact distance at the vertex where the contact state is detected.
[0125] Changes in contact distance are combined with the coefficient of restitution based on the material to generate a repulsive force vector, and the contact pressure can be derived by converting the magnitude of the vector into units of contact area.
[0126] For example, if the waistband area strongly compresses the waist cross-section of the 3D avatar, a pressure of 3.5 kPa or more may be produced at the mesh vertices of the 3D clothing in that area, and this may be displayed in red on the simulation results.
[0127] Contact pressure information is used for evaluating wear discomfort, displaying compression intensity, and simulating material correction, and areas exceeding a certain standard can be set as targets for automatic notifications or design feedback.
[0128] The simulation unit (140) can divide the generated simulation information into mesh units of 3D clothing and organize it into structured simulation data.
[0129] Simulation information may include multidimensional vector information including elongation rate, curvature, and contact pressure values, which can be transmitted to a rendering unit (150) to be described later and used for visualization calculations.
[0130] For example, when a user performs the motion of raising their arm, if a 12% elongation rate is assigned to the shoulder mesh vertices of the 3D garment, a curvature of 1.5 kPa is assigned to the elbow vertices, and a contact pressure of 3.8 kPa is assigned to the underarm vertices, this information is visualized in the form of a real-time color map so that the wearer can intuitively check the garment's response to the wearer's movement.
[0131] Meanwhile, the simulation unit (140) can generate prediction simulation information that reflects future body shape changes of the 3D avatar based on additional user information received.
[0132] The simulation unit (140) can predict the body shape of a 3D avatar up to a temporally extended point in time by linking with an artificial intelligence prediction module to additionally provide the user with body shape changes and fitting results according to the user's health condition, physiological changes or environmental factors.
[0133] The simulation unit (140) can quantitatively simulate the deformation response of a 3D garment considering the possibility of a user's actual body shape change, and the predicted result can be reflected in the elongation rate, curvature, and contact pressure of the 3D garment and provided as a prediction simulation result for future situations.
[0134] User information may include health status information, age information, weight change trend information, pregnancy status information, hormonal change history information, and seasonal environmental information, and may be manually entered through a user interface or automatically collected from past wearing records, biosensor devices, personal health record APIs, local environment databases, etc.
[0135] For example, if records of continuously increasing abdominal contact pressure are collected through recent wearing history, or if the average number of steps from the wearable device decreases, the simulation unit (140) can determine that the user's body shape is on an increasing trend based on this data and generate a 3D avatar that reflects this.
[0136] Additionally, for female users, if pregnancy status and week information are manually entered or received via integration with an external app, body shape changes based on the week of pregnancy can be automatically applied.
[0137] For example, if the user is “6 months pregnant,” the simulation unit (140) can generate a 3D avatar body shape at the 8th month of pregnancy based on data predicting changes in the abdominal expansion, hips, and thighs according to the progression of pregnancy, and can pre-calculate simulation information for the elongation rate, curvature, and contact pressure of the 3D clothing that fits it. Afterward, the user can predict in advance whether the clothing will still be wearable in the body shape 2 months later, and whether discomfort will occur in specific areas.
[0138] Seasonal information is collected through the system's calendar and location-based APIs, and, for example, body types reflecting a tendency for reduced activity and increased body fat during the winter season can be predicted.
[0139] For example, during the winter, changes in body shape may occur in the abdomen and upper body due to an increase in average body weight and reduced outdoor activity; conversely, during the summer, the body may shift toward requiring a thin fit to enhance the efficiency of sweat wicking and heat dissipation. These seasonal changes can be reflected in simulation standards regarding clothing breathability, pressure relief in tight areas, and prevention of wrinkle formation.
[0140] Additionally, the simulation unit (140) may include changes in body shape and activity levels according to the season as predictive factors. For example, in winter, changes in body shape may occur in the abdomen and upper body due to an increase in average body weight and a decrease in outdoor activity, while conversely, in summer, the body may change in a direction that requires a thin fit to increase the efficiency of sweat release and heat dissipation. These seasonal changes can be reflected as simulation reference values for ensuring breathability of the clothing, relieving pressure on close-fitting areas, and preventing wrinkle formation.
[0141] The artificial intelligence prediction module may be composed of a recurrent neural network (RNN) with a Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU) structure that performs time-series-based body shape change modeling.
[0142] The input to the artificial intelligence prediction module is a sequence vector in which numerical information such as age, weight, gestational week, hormonal change indicators, and activity level is arranged in chronological order, and the prediction result may consist of body shape measurements such as height, chest circumference, waist circumference, hip circumference, pelvic tilt, and trunk curvature at a future point in time.
[0143] The simulation unit (140) can transmit the prediction results to the avatar generation unit (120) to regenerate a 3D avatar, and receive the regenerated 3D avatar back to simulate the fitting of the 3D clothing.
[0144] Through this, the simulation unit (140) can provide prediction-based virtual fitting results that reflect not only the current body shape but also the possibility of future body changes, and can realize a user-customized simulation function.
[0145] The rendering unit (150) visually outputs a 3D avatar fitted with a 3D outfit based on simulation information received from the simulation unit (140), and is electrically connected to the avatar generation unit (120), the outfit generation unit (130), and / or the simulation unit (140).
[0146] The rendering unit (150) can perform a rendering operation function that converts simulation information into a real-time renderable format to output the fitting status and dynamic reaction of the 3D clothing to a visual screen.
[0147] The rendering unit (150) can construct a rendering scene based on simulation information received from the simulation unit (140).
[0148] A rendering scene is defined as a virtual 3D space containing 3D avatars and 3D clothing, and this space represents the basic unit for performing rendering operations.
[0149] The rendering scene may include visual elements such as the vertex structure of the 3D avatar, the mesh structure of the 3D clothing, mesh deformation information, light source position, virtual camera viewpoint, light reflection and / or shadow calculation conditions.
[0150] A rendering scene represents a structure that integrates all visual elements to visually represent simulation results, with 3D objects placed and light sources and viewpoints defined.
[0151] The rendering unit (150) periodically updates the mesh vertex positions of the 3D garment along the simulation time axis and can reconstruct the deformation state of the 3D garment within the rendering scene by reflecting the vertex positions updated in each time frame.
[0152] Each frame of the rendering scene can precisely reflect the state of the 3D clothing based on the simulation results and can be displayed on the user screen through GPU-based real-time computation.
[0153] This rendering scene configuration method enables users to intuitively recognize the changing state of 3D clothing by implementing visual fitting results similar to the actual physical environment.
[0154] The rendering unit (150) can update the deformation data of the 3D avatar and 3D clothing in real time at each frame of the simulation time axis and write the rendering operation result to the frame buffer and output it to the user interface.
[0155] Rendering resolution, shading method, and mesh wireframe status are automatically adjusted based on user settings or system performance conditions, and high-speed rendering can be performed through GPU-based hardware acceleration.
[0156] The rendering unit (150) can support various visualization options to precisely analyze the fitting results of the 3D clothing.
[0157] Users can freely adjust the virtual camera position to observe the fitting status of the 3D garment from any desired viewpoint, such as the upper body, lower body, side, or back, and can continuously check the changes in the responsiveness of the 3D garment frame by frame through the time axis control function.
[0158] In addition, by utilizing the mesh transparency adjustment function, you can clearly identify the internal wrinkle structure or the relative positional relationship with the 3D avatar.
[0159] The rendering unit (150) can independently process vertex unit data for each of the 3D avatar, 3D clothing, and simulation information during all visual output processes, and can provide the user with fitting results similar to actual wearing in real time.
[0160] In addition, the rendering unit (150) can visualize the 3D clothing fitting results at a future point in time based on the prediction simulation information.
[0161] The prediction simulation information includes the fitting status, elongation rate, curvature, and contact pressure information of the future body shape-based 3D avatar and the corresponding 3D clothing generated by the simulation unit (140), and can be allocated to a separate time interval within the rendering scene and output.
[0162] The rendering unit (150) can compare a rendering frame at the current time and a predicted rendering frame at a future time and output them simultaneously, or visually highlight and display the difference between the time points selected by the user.
[0163] For example, simulation results of wearing the same 3D clothing in the current body shape and the body shape after 8 months of pregnancy are sequentially displayed in the rendering scene, and the elongation rate, curvature, and contact pressure at each point in time can be output as dynamic visual information in real time.
[0164] Additionally, the rendering unit (150) can provide a function to highlight the difference in deformation between time points for comparative analysis of the prediction simulation results.
[0165] For example, when comparing the fitting results of a summer body type and a winter body type for the same outfit, the rendering unit (150) can enlarge the mesh of the area with large deformation or apply color contrast to make it easier for the user to visually recognize.
[0166] Additionally, the rendering unit (150) can place the generated rendering scenes on the timeline according to the predicted time points and support the user in exploring the predicted scenarios in a time-series manner through a time-shift control function.
[0167] Mesh deformation information of the 3D clothing included in the prediction simulation information is updated in real time according to the body shape conditions at each point in time and can be displayed on the screen without time delay through a GPU-based rendering engine.
[0168] This method allows the user to check in advance the suitability for wearing, the risk of compression, and the possibility of wrinkles due to changes in body shape in the present as well as the future, and enables the prediction simulation results to be delivered intuitively and realistically through the rendering unit (150).
[0169] As described above, according to the artificial intelligence-based virtual clothing fitting system (100) of one embodiment of the present invention, digital clothing can be virtually fitted to a 3D avatar generated based on the user's body information, and the results can be visually provided while simultaneously quantitatively analyzing fitting responses such as deformation patterns or tightness characteristics of the clothing. Therefore, a fitting experience close to actual wearing can be provided in an online environment, and consumers can more accurately select clothing suitable for their body shape, thereby reducing the return rate and increasing purchase satisfaction.
[0170] In addition, various functions such as size recommendation, fit prediction, and wear simulation can be automated based on the results of quantitative analysis of wear response. This can be utilized in various fields such as clothing design, virtual fitting shopping malls, and personalized recommendation systems, thereby contributing to the improvement of user experience as well as distribution efficiency and the digital transformation of the entire industry.
[0172] Although all components constituting the embodiments of the present invention have been described above as being combined or operating in combination, the present invention is not necessarily limited to such embodiments. That is, within the scope of the purpose of the present invention, all components may be selectively combined in one or more ways to operate.
[0173] Furthermore, terms such as "include," "compose," or "have" described above, unless specifically stated otherwise, mean that the relevant component may be inherent; therefore, they should be interpreted as allowing for the inclusion of additional components rather than excluding them. All terms, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains, unless otherwise defined. Commonly used terms, such as those defined in advance, should be interpreted in accordance with their meaning in the context of the relevant technology and should not be interpreted in an ideal or overly formal sense unless explicitly defined in the present invention.
[0174] Furthermore, the above description is merely an illustrative explanation of the technical concept of the present invention, and those skilled in the art to which the present invention pertains will be able to make various modifications and variations within the scope of the essential characteristics of the present invention.
[0175] Accordingly, the embodiments disclosed in this invention are intended to illustrate, not limit, the technical concept of the invention, and the scope of the technical concept of the invention is not limited by these embodiments. The scope of protection of this invention shall be interpreted by the claims below, and all technical concepts within an equivalent scope shall be interpreted as being included within the scope of rights of this invention. Explanation of the symbols
[0176] 100: Artificial intelligence-based virtual clothing fitting system according to an embodiment of the present invention 110: Body Information Generation Unit 111: Preprocessing module 112: Extraction Module 113: Operation Module 120: Avatar Creation Section 130: Costume Creation Section 140: Simulation section 150: Rendering section P: Feature point
Claims
Claim 1 A body information generation unit that generates body information including the user's height, chest circumference, waist circumference, and hip circumference based on a body image of the user captured by a 3D camera; an avatar generation unit that generates a 3D avatar based on the body information; a garment generation unit that generates a 3D garment based on a previously stored garment drawing; and a simulation unit that fits the 3D garment to the 3D avatar to generate simulation information indicating the elongation rate, curvature, and contact pressure of the 3D garment. The system includes a rendering unit that visually outputs the 3D avatar fitted with the 3D clothing based on the simulation information, and the body information generation unit includes a preprocessing module that generates 3D point cloud data from the body image and preprocesses the body image by performing background removal, noise filtering, surface refinement, and alignment processing on the point cloud data; an extraction module that inputs the preprocessed body image into an artificial intelligence model based on a PointNet architecture to extract feature points from the point cloud data, wherein if the prediction reliability of the feature points falls below a preset threshold, the feature points are discarded or classified as targets for re-estimation; and a computation module that generates the body information by performing a consistency check by comparing the distance between the feature points with standard human body ratios, determining feature points among the feature points that exceed a preset tolerance as errors, and calculating the user's height, chest circumference, waist circumference, and hip circumference based on the distance between feature points that passed the consistency check. The simulation unit calculates the elongation rate, the curvature, and the contact pressure by placing the 3D clothing on the surface of the 3D avatar, wherein the elongation rate is the 3D Calculated based on the degree of surface deformation of the garment, the curvature is calculated based on the change in curved surface inclination and folding distribution of the 3D garment, the contact pressure is calculated based on the change in contact position and contact distance between the 3D garment and the 3D avatar surface, and the simulation unit includes health status information for the user,Predictive simulation information reflecting future body shape changes of the 3D avatar is generated through an artificial intelligence prediction module comprising a recurrent neural network (RNN) of a Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU) structure that performs time-series-based body shape change modeling based on user information including age information, weight change trend information, pregnancy status information, gestational week information, hormone change history information, and seasonal environmental information; wherein, if the user is pregnant, the artificial intelligence prediction module generates a 3D avatar at a future point in time by reflecting the body shape changes of the user's abdomen, hips, and thighs according to the gestational week information, and generates the predictive simulation information by calculating the elongation rate, curvature, and contact pressure of the 3D clothing for the 3D avatar at the future point in time; wherein, based on the seasonal environmental information, the 3D avatar at the future point in time is generated by reflecting body shape changes of the abdomen and upper body due to reduced outdoor activity and weight gain in winter, and body shape reduction due to increased sweat elimination and heat dissipation efficiency in summer, and the 3D clothing for the 3D avatar at the future point in time An AI-based virtual clothing fitting system that calculates elongation rate, curvature, and contact pressure to generate the prediction simulation information, and the rendering unit generates a rendering frame at a current time point based on the simulation information and generates a rendering frame at a future time point based on the prediction simulation information, and simultaneously outputs the rendering frame at the current time point and the rendering frame at the future time point, while applying enlargement or color contrast to mesh areas where the difference in elongation rate, curvature, and contact pressure between time points exceeds a preset reference value. Claim 2 delete Claim 3 An artificial intelligence-based virtual clothing fitting system according to claim 1, wherein the avatar generation unit generates a 3D avatar whose shape can be adjusted according to the body information through a parametric human body model. Claim 4 An artificial intelligence-based virtual clothing fitting system according to claim 1, wherein the clothing generation unit generates 3D panels for each part of the clothing based on the clothing drawing, and generates the 3D clothing by combining a plurality of the 3D panels according to the seam lines of the clothing indicated on the clothing drawing. Claim 5 delete