AI-based integrated clothing customization system and method

Abstract

This invention provides an integrated clothing customization system and method based on AI technology, belonging to the field of smart retail and e-commerce technology. The system adopts a dual-space architecture: the first space enables contactless AI body measurement, 3D human body modeling, and digital interaction; the second space automatically handles the storage and retrieval of physical sample garments. It integrates modules for size extraction, virtual try-on rendering, physical try-on feedback, and order generation. AI algorithms achieve accurate body measurement, simultaneous virtual and physical try-on, and parametric pattern generation. The method includes human body image acquisition, 3D modeling and size extraction, virtual try-on, physical try-on feedback, order generation, and production integration. This invention achieves integrated measurement, try-on, customization, and production processes, improving measurement accuracy and the try-on experience while shortening the production cycle.

Description

Technical Field

[0001] This invention relates to the field of smart retail and e-shopping technology, and in particular to an integrated clothing customization system and method based on AI technology. Background Technology

[0002] With the popularization of the Internet, big data and mobile payment, e-shopping has become one of the mainstream shopping methods for consumers around the world. Among them, clothing, as the largest retail e-commerce category, has undergone a development process in its online sales model, from the initial text and image display to model video display, and then to the virtual fitting technology that has emerged in recent years.

[0003] In the early stages of e-commerce development, consumers could only rely on flat lay or model images and size charts to make purchases. Due to a lack of accurate understanding of their own body shape and the fit of clothing, consumers often made blind purchases, leading to high return rates due to ill-fitting items – a major pain point for the industry. To address this issue, the industry developed virtual fitting technology based on augmented reality (AR) or 3D modeling. Initial solutions typically required users to manually input data such as height, weight, and bust, waist, and hip measurements. The system would then generate an approximate human body model for a fitting simulation. However, manual measurements had significant errors, and ordinary users lacked professional measurement skills, resulting in a noticeable discrepancy between the virtual model and the user's actual body shape, limiting the fitting's reference value. Subsequently, attempts emerged to use smartphone monocular cameras for 3D human body reconstruction, but its accuracy was limited by ambient lighting, clothing thickness, and the phone's processing power, making it difficult to achieve the millimeter-level precision required for custom clothing.

[0004] In recent years, brick-and-mortar stores have begun to introduce smart body measurement devices with depth cameras, which can accurately obtain users' body shape data. However, existing solutions still have obvious problems with fragmented experience: although the body measurement, virtual selection, and physical sample fitting confirmation are integrated in the same store, the physical samples still need to be retrieved from the warehouse by the store staff, requiring users to wait multiple times. Moreover, the data from virtual fitting and physical fitting cannot be automatically synchronized, making it impossible to form a coherent consumption decision.

[0005] Furthermore, the existing system has gaps in the connection between the backend and the manufacturing process. Users' personalized modification requests are mostly transmitted to the factory in the form of text notes, and then pattern makers manually modify the CAD drawings based on experience. This unstructured data transmission method is prone to misunderstandings, resulting in customized products that do not meet user expectations. It also prolongs production preparation time and restricts the application of personalized clothing customization.

[0006] Therefore, how to provide a complete front-end solution that integrates high-precision non-contact measurement, AI intelligent interaction, real-time realistic virtual rendering, automatic storage and retrieval of physical samples, and direct connection to parametric production is a key challenge that urgently needs to be overcome in the current cross-disciplinary technology field of e-commerce and clothing customization. Summary of the Invention

[0007] This invention provides an integrated clothing customization system and method based on AI technology to solve the aforementioned technical problems.

[0008] This invention provides an integrated clothing customization system based on AI technology, comprising: The first and second spaces are interconnected; The first space is equipped with a mirror display screen, a body measurement data acquisition module, and a user session management module. The body measurement data acquisition module is used to collect multi-angle body image data of users, and the user session management module is used to realize multi-user data isolation and queuing management. The second space is used to store physical sample garments of various models and styles, and to unlock the electromagnetic electric lock of the corresponding model sample garment according to the user's confirmed fitting instructions. The size extraction module, connected to the body measurement data acquisition module, is used to construct a 3D digital human body model of the user based on the multi-angle body image data, and extract key human body size data through the optimal circumference cross-section model, circumference curve closed length model and surface correction model; the digital interaction module, connected to the size extraction module and the mirror display screen, is used to generate and display the user's 3D digital human body model, and recommend clothing parameters based on the user's clothing preferences. The virtual fitting rendering module, connected to the digital interaction module, is used to predict the circumference after wearing the garment based on the key human body size data and the clothing parameters selected by the user, combined with the physical properties of the fabric, and drive the clothing deformation mesh to generate the virtual fitting effect. The physical try-on feedback module is connected to the mirror display screen and the digital interaction module. It is used to receive feedback from users on the adjustment of the circumference and non-circumference parameters of the physical try-on sample garment, generate a secondary adjustment vector through the physical-virtual difference quantization algorithm, and synchronously update the 3D digital human body model and the virtual try-on effect. The order generation module, connected to the digital interaction module and the physical try-on feedback module, is used to generate a customized order containing structured pattern parameters based on the clothing parameters finally confirmed by the user, and to provide order modification and cancellation functions.

[0009] Preferably, the first space is an independent, enclosed compartment, and an automatic sensor door is installed between it and the second space; A circular standing sign is placed in the center of the ground in the first space, and depth cameras are evenly installed on the surrounding walls, divided into upper and lower layers, with LED soft lights installed on the top.

[0010] Preferably, the body measurement data acquisition module includes an FPGA synchronous controller and a guidance unit; The FPGA synchronization controller enables synchronous triggering of the camera; The guidance unit guides users to complete standard postures through digital human animation and voice prompts, and uses a human key point detection algorithm to detect postures in real time. When all joint angles meet the standard, it automatically triggers shooting.

[0011] Preferably, the size extraction module has a built-in curvature outlier filtering algorithm, which considers points with curvature exceeding three times the standard deviation of the average value as outliers and removes them before calculating the surface correction amount; At the same time, the human body is divided into two halves, left and right, and the optimal circumference cross section of the left and right halves is calculated separately. When the deviation of the left and right dimensions exceeds the threshold, the user is prompted on the interactive interface and the left and right dimensions are displayed respectively.

[0012] Preferably, the digital interaction module calculates the sample garment matching score through a matching model, where the matching model is: ; in, For the first The matching score of the sample garment stored in the second space; The total number of dimensions involved in the matching; For user number The final dimensions of each circumference; For the first Sample garment The original net dimensions of each circumference; For the first The preset optimal ease for each circumference; For the first Tolerance coefficient for each dimension; Fabric attribute matching factor; Adjustments based on user clothing preferences.

[0013] Preferably, the virtual fitting rendering module uses a linear hybrid skinning algorithm combined with position-based dynamics to simulate clothing deformation and wrinkles; The girth prediction model after wearing is as follows: ; in, Predicted circumference after wearing; The basic relaxation constant; The elastic modulus of the fabric for the selected physical sample garment. For fabric thickness, For fabric surface density, The characteristic length of the fabric stretch. This is the proportionality coefficient.

[0014] Preferably, the second space is a modular storage area with multiple expandable storage compartments, each equipped with an electromagnetic electric latch, an infrared presence detection sensor, an RFID reader, and an LED indicator.

[0015] Preferably, the interactive interface of the physical try-on feedback module supports the adjustment of eight parameters: bust, waist, hip, shoulder width, garment length, sleeve length, collar height, and trouser length. The entity-virtual difference quantization algorithm is as follows: in, , These are preset visual weights and preset tactile weights, which are pre-set by the system according to the clothing category; This represents the secondary adjustment amount for the k-th part; Adjustments based on the amount of physical fitting input by the user; Mapping values ​​for the tightness level input by the user; This is the tightness adjustment range coefficient.

[0016] Preferably, the final version parameter vector model is as follows: ;in, For Hadama accumulation, It is a vector of all 1s. This is the overall scaling factor. To prevent division by zero minimum constant, For the preset dimensional weight vector Based on the basic pattern parameter vector, This is the final girth dimension vector confirmed by the user. This is a vector for correcting fabric properties. This is the parameter vector for the final version.

[0017] This invention provides an integrated clothing customization method based on AI technology, comprising: Collect multi-angle body image data of users; A 3D digital human body model of the user is constructed based on the multi-angle body image data, and key human body size data is extracted through the optimal circumference section model, circumference curve closed length model and surface correction model. Based on the key human body size data and the clothing parameters selected by the user, the circumference after wearing is predicted by combining the physical properties of the fabric, and the clothing deformation mesh is driven to generate a virtual fitting effect. The system receives user feedback on adjustments to the circumference and non-circumference parameters of the physical try-on sample garment, generates a secondary adjustment vector through a physical-virtual difference quantization algorithm, and synchronously updates the 3D digital human body model and the virtual try-on effect. A customized order containing structured pattern parameters is generated based on the user's final confirmed clothing parameters, and the order modification and cancellation functions are provided. Compared with the prior art, the beneficial effects of this application are as follows: 1. Significantly improved measurement accuracy: By using multi-view 3D camera acquisition combined with a three-level size extraction algorithm of "optimal cross-section adaptive selection + curvature anomaly point filtering + left and right half separation measurement", the average measurement error is controlled within 0.8mm, which can meet the millimeter-level accuracy requirements of most clothing customization scenarios. 2. Seamless and smooth fitting experience: Adopting a dual-space integrated architecture combined with automatic sample garment storage and retrieval, users do not need to wait for store staff to retrieve the garment; through the physical-virtual difference quantification algorithm, virtual and physical fitting data are synchronized, reducing the average number of fittings per user and shortening the time for customized services per customer; 3. Accurate and efficient data flow: The user's subjective adjustment opinions are transformed into structured pattern parameters, which are directly read by the factory CAD system, eliminating the misunderstanding of text notes and shortening the pattern design and paper preparation time from the traditional 1-2 days to less than 1 hour. 4. Strong system robustness: The system has added modules for user session management, sample garment lifecycle management and exception handling, supports multi-user queuing, offline working mode and hardware failure degradation handling, and can be stably applied to actual commercial scenarios such as shopping malls and stores.

[0018] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural diagram of an AI-based integrated clothing customization system according to an embodiment of the present invention; Figure 2 This is a flowchart of an AI-based integrated clothing customization method according to an embodiment of the present invention. Detailed Implementation

[0021] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0022] This invention provides an integrated clothing customization system based on AI technology, such as... Figure 1 As shown, it includes: The first and second spaces are interconnected; The first space is equipped with a mirror display screen, a body measurement data acquisition module, and a user session management module. The body measurement data acquisition module is used to collect multi-angle body image data of users, and the user session management module is used to realize multi-user data isolation and queuing management. The second space is used to store physical sample garments of various models and styles, and to unlock the electromagnetic electric lock of the corresponding model sample garment according to the user's confirmed fitting instructions. The size extraction module, connected to the body measurement data acquisition module, is used to construct a 3D digital human body model of the user based on the multi-angle body image data, and extract key human body size data through the optimal circumference cross-section model, circumference curve closed length model and surface correction model; the digital interaction module, connected to the size extraction module and the mirror display screen, is used to generate and display the user's 3D digital human body model, and recommend clothing parameters based on the user's clothing preferences. In this embodiment, the optimal girth section model refers to a mathematical model that adaptively selects the smoothest and most representative horizontal section of the corresponding part of the human body by minimizing the cross-sectional curvature variance, in order to avoid non-feature sections caused by posture offset and scanning noise. The girth curve closure length model refers to a mathematical model that sorts the point cloud on the optimal section according to the polar angle, calculates the sum of the Euclidean distances between adjacent points, and obtains the initial girth length. The surface correction model refers to a mathematical model that corrects the initial girth length based on the deviation between the local curvature of the cross-sectional point cloud and the standard reference curvature, thereby obtaining the final accurate dimensions.

[0023] The virtual fitting rendering module, connected to the digital interaction module, is used to predict the circumference after wearing the garment based on the key human body size data and the clothing parameters selected by the user, combined with the physical properties of the fabric, and drive the clothing deformation mesh to generate the virtual fitting effect. The physical try-on feedback module is connected to the mirror display screen and the digital interaction module. It is used to receive feedback from users on the adjustment of the circumference and non-circumference parameters of the physical try-on sample garment, generate a secondary adjustment vector through the physical-virtual difference quantization algorithm, and synchronously update the 3D digital human body model and the virtual try-on effect. In this embodiment, the entity-virtual difference quantization algorithm refers to the algorithm that weights and quantifies the user's visual adjustment feedback and tactile tightness feedback on the physical sample garment to generate a secondary adjustment vector for the 3D digital human body model.

[0024] The order generation module, connected to the digital interaction module and the physical try-on feedback module, is used to generate a customized order containing structured pattern parameters based on the clothing parameters finally confirmed by the user, and to provide order modification and cancellation functions.

[0025] Preferably, the first space is an independent, enclosed compartment, and an automatic sensor door is installed between it and the second space; A circular standing sign is placed in the center of the ground in the first space, and depth cameras are evenly installed on the surrounding walls, divided into upper and lower layers, with LED soft lights installed on the top.

[0026] Preferably, the body measurement data acquisition module includes an FPGA synchronous controller and a guidance unit; The FPGA synchronization controller enables synchronous triggering of the camera; The guidance unit guides users to complete standard postures through digital human animation and voice prompts, and uses a human key point detection algorithm to detect postures in real time. When all joint angles meet the standard, it automatically triggers shooting.

[0027] Preferably, the size extraction module has a built-in curvature outlier filtering algorithm, which considers points with curvature exceeding three times the standard deviation of the average value as outliers and removes them before calculating the surface correction amount; At the same time, the human body is divided into two halves, left and right, and the optimal circumference cross section of the left and right halves is calculated separately. When the deviation of the left and right dimensions exceeds the threshold, the user is prompted on the interactive interface and the left and right dimensions are displayed respectively.

[0028] Preferably, the digital interaction module calculates the sample garment matching score through a matching model, where the matching model is: ; in, For the first The matching score of the sample garment stored in the second space; The total number of dimensions involved in the matching; For user number The final dimensions of each circumference; For the first Sample garment The original net dimensions of each circumference; For the first The preset optimal ease for each circumference; For the first Tolerance coefficient for each dimension; Fabric attribute matching factor; Adjustments based on user clothing preferences.

[0029] In this embodiment, the total number of dimensions involved in the matching is dynamically adjusted according to the clothing category: for tops such as suits, T-shirts, dresses, and coats: K=4 (chest, waist, hip, shoulder width); for bottoms such as jeans and casual pants: K=3 (waist, hip, pant length); for shirts: K=5 (chest, waist, hip, shoulder width, sleeve length).

[0030] It should be noted that, By fusing size deviation, ease, preference and fabric attributes with Gaussian kernel, the matching results are normalized to the [0,1] interval, realizing AI automatic and accurate recommendation of sample garments. This solves the problem of large deviation in traditional recommendations and the need for manual screening, and is the core basis for automatic linkage between the two spaces.

[0031] Preferably, the virtual fitting rendering module uses a linear hybrid skinning algorithm combined with position-based dynamics to simulate clothing deformation and wrinkles; The girth prediction model after wearing is as follows: ; in, Predicted circumference after wearing; The basic relaxation constant; The elastic modulus of the fabric for the selected physical sample garment. For fabric thickness, For fabric surface density, The characteristic length of the fabric stretch. This is the proportionality coefficient.

[0032] In this embodiment, The value is The calibration was obtained through 100 sets of physical fitting experiments with different fabrics.

[0033] It should be noted that the nonlinear girth prediction model built based on the physical properties of the fabric conforms to the deformation law of the fabric and can accurately calculate the virtual wearing girth. This solves the pain point of virtual try-on lacking physical calculation and being disconnected from physical try-on, and provides benchmark data for virtual-real synchronization.

[0034] Preferably, the second space is a modular storage area with multiple expandable storage compartments, each equipped with an electromagnetic electric latch, an infrared presence detection sensor, an RFID reader, and an LED indicator.

[0035] Preferably, the interactive interface of the physical try-on feedback module supports the adjustment of eight parameters: bust, waist, hip, shoulder width, garment length, sleeve length, collar height, and trouser length. The entity-virtual difference quantization algorithm is as follows: in, , These are preset visual weights and preset tactile weights, which are pre-set by the system according to the clothing category; This represents the secondary adjustment amount for the k-th part; Adjustments based on the amount of physical fitting input by the user; Mapping values ​​for the tightness level input by the user; This is the tightness adjustment range coefficient.

[0036] In this embodiment, 1 / 10 represents the tightness adjustment range coefficient. The unit is converted from mm to cm.

[0037] It should be noted that by using a dual weighting of visual and tactile feedback, the user's subjective fitting feedback is quantified into size adjustments, which aligns with the logic of human body fitting evaluation. This enables automatic synchronization of virtual and real fitting data, solving the problems of errors in manual annotations and the inability to drive model updates.

[0038] Preferably, the final version parameter vector model is as follows: ;in, For Hadama accumulation, It is a vector of all 1s. This is the overall scaling factor. To prevent division by zero minimum constant, For the preset dimensional weight vector The basic pattern parameter vector, This is the final girth dimension vector confirmed by the user. This is a vector for correcting fabric properties. This is the parameter vector for the final version.

[0039] In this embodiment, the overall scaling factor is set to 0.8, which is obtained by training with pattern data from 500 sets of customized orders. This balances the flexibility and stability of pattern adjustment and avoids pattern distortion caused by excessive adjustment.

[0040] It should be noted that the use of Hadamard product and tanh normalization to achieve element-by-element pattern correction integrates basic pattern, circumference weight and fabric properties, conforms to the logic of industrial pattern making, and can generate structured parameters that can be read directly from CAD, solving the problems of inefficiency and easy deviation of manual pattern modification.

[0041] The first space is a separate physical compartment for user measurement, virtual interaction, and fitting feedback. It integrates measurement data acquisition equipment, a mirrored display screen, and auxiliary facilities, and is the core area for digital interaction between the user and the system. The first space is a 2m (length) × 2m (width) × 2.2m (height) independent enclosed compartment with white matte walls to reduce light reflection. The layout of the space is as follows: A circular standing marker with a diameter of 50cm is placed in the center of the ground, with the center of the marker indicating the best standing point for the user; A 100-inch mirrored display screen is installed on the front wall, which combines the functions of a mirror and a display, and is used to display 3D digital human body models, virtual fitting effects and interactive interfaces; Eight Intel RealSense D455 depth cameras are evenly installed on the four walls, in two layers: the lower layer has four cameras installed at a height of 1.2m (corresponding to the middle of the human torso), and the upper layer has four cameras installed at a height of 1.7m (corresponding to the shoulders of the human body). A 1-meter-wide automatic sensor door is installed on the right wall to connect the first space and the second space; A locker is set up in the corner to store black Lycra bodysuits in sizes S, M, L, and XL, so that users can change into them to reduce the interference of clothing wrinkles on measurements; Four sets of LED soft lights are installed on the top to provide a uniform, shadow-free lighting environment, with the light intensity controlled between 500 and 800 lux.

[0042] The second space, physically connected to the first space, serves as a storage and retrieval area for physical try-on garments. An automated electronic control system enables intelligent unlocking and return of the garments, acting as a physical bridge between virtual and physical try-on. Adjacent to the right of the first space is a 1.5m (length) × 3m (width) × 2.2m (height) modular storage area, connected to the first space via an automatic sensor door. This area contains 40 independent storage compartments arranged in 5 columns × 8 rows, each measuring 30cm (width) × 40cm (depth) × 60cm (height), constructed of cold-rolled steel with a powder-coated surface. The storage compartments can be modularly expanded to a maximum of 120 compartments in 15 columns × 8 rows. Each compartment is equipped with: Electromagnetic electric latch, operating voltage 12V, response time ≤0.5s; An infrared in-situ detection sensor is used to detect whether the sample garment is within the grid. RFID reader / writer, used to read RFID tags inside the sample garment; LED indicator lights: green indicates available, red indicates locked, and yellow indicates disinfection required. In this embodiment, all modules in the system are connected via a gigabit LAN, using the Transmission Control Protocol / Internet Protocol (TCP / IP) communication protocol, and the data transmission format is JavaScript Object Notation (JSON). The data flow logic is as follows: The user session management module creates user sessions and assigns unique user IDs. The body measurement data acquisition module sends the acquired multi-angle body images to the size extraction module; The size extraction module processes and generates a 3D digital human body model and key size data, and sends them to the digital interaction module and the virtual fitting rendering module. The digital interaction module generates sample garment recommendations based on size data and user preferences, and sends them to the control unit and virtual fitting rendering module in the second space; The virtual try-on rendering module generates static and dynamic virtual try-on effects and sends them to the mirror display screen for display. The physical try-on feedback module receives adjustment information input by the user, generates a secondary adjustment vector, and sends it to the digital interaction module; The digital interaction module updates the 3D digital human body model, triggering the virtual try-on rendering module to re-render. The order generation module summarizes the parameters confirmed by the user, generates a customized order, and sends it to the factory manufacturing execution system (MES) and the parametric garment CAD system; The sample garment lifecycle management module obtains sample garment information through RFID readers, updates inventory status, and triggers disinfection reminders. The exception handling module monitors the operating status of each module in real time, and triggers an alarm and initiates degradation processing when an exception occurs.

[0043] In this embodiment, the hardware parameters of the body measurement data acquisition module are as follows: 3D camera: Uses Intel RealSense D455 depth camera, resolution 1920×1080, depth accuracy ±2mm@2m, frame rate 30 frames / second (fps), field of view 87°×58°; FPGA Synchronization Controller: Employs a Xilinx Spartan-6 Field Programmable Gate Array (FPGA) to achieve synchronous triggering of 8 cameras with a triggering accuracy of ≤1ms; The guidance unit is integrated into the mirrored display screen, guiding users to complete standard postures through digital human animation and voice prompts.

[0044] In this embodiment, the standard pose quantification standard is: The guidance unit directs users to maintain the following standard posture: Torso: Keep naturally upright, with the back ≤5cm from the wall, shoulders relaxed and lowered, and head looking straight ahead; Upper limbs: Arms hang naturally, forming an angle of 15°±5° with the torso, hands naturally clenched into fists, palms facing the body; Lower limbs: Stand with your legs shoulder-width apart (about 30cm ± 5cm), toes pointing forward, and knees naturally straight.

[0045] In this embodiment, the system uses a human key point detection algorithm to identify 21 key skeletal points of the user (nose, shoulder, elbow, wrist, hip, knee, ankle, etc.) in real time and calculate the angles of each joint: When the torso tilts forward or backward at an angle greater than 5° or tilts left or right at an angle greater than 3°, a voice prompt will say, "Please straighten your back and keep your body upright." When the angle of your arms is less than 10° or greater than 20°, the voice prompt will say, "Please place your arms naturally at your sides, keeping a fist's distance from your body." When the distance between your legs is less than 25cm or greater than 35cm, the voice prompt will say, "Please separate your feet to shoulder width." Once all joint angles meet the standard and remain stable for 2 seconds, the system automatically triggers 8 cameras to simultaneously capture images of the body from 8 angles.

[0046] In this embodiment, the point cloud preprocessing steps of the size extraction module are as follows: Before extracting dimensions, the system first preprocesses the collected point cloud data: Statistical filtering: Calculate the average distance from each point to its 50 nearest neighbors and remove outliers whose distances are greater than three times the standard deviation of the average distance; Radius filtering: Points with fewer than 5 neighboring points within a 5mm radius of each point are considered noise points and removed; Poisson reconstruction: The preprocessed point cloud data is converted into a triangular mesh model, with a reconstruction accuracy of 0.5mm; Facial blurring: Automatically identifies and blurs facial features of 3D models to protect user privacy.

[0047] In this embodiment, the optimal girth section model is as follows: For different girth locations, the system generates candidate sections within the corresponding height range: Chest circumference: Within a height range of 5cm from the armpit to the nipple, a horizontal candidate section is generated every 2mm, for a total of 25 sections; Waist circumference: Within a height range of 5cm above and below the navel, a horizontal candidate section is generated every 2mm, for a total of 50 sections; Hip circumference: Within a height range of 5cm above and below the most prominent part of the hip, a horizontal candidate section is generated every 2mm, for a total of 50 sections; Shoulder width: Within a height range of 2cm above and below the acromion, a horizontal candidate section is generated every 1mm, for a total of 40 sections.

[0048] In this embodiment, the optimal cross section is selected using an optimal girth cross section model, which is as follows: ,in, For the first The optimal cross-sectional curve for an individual's body circumference; For the first The first dimension Candidate cross-sectional curves; Candidate cross section The number of 3D point clouds on the surface; cross section Upper A 3D point coordinate vector; for The local surface curvature; cross section The average curvature of all points on it; This model adaptively selects the smoothest cross section as the optimal girth cross section by minimizing the cross section curvature variance, thus avoiding non-feature cross sections caused by pose offset and scan noise.

[0049] In this embodiment, the girth curve closure length model is used to calculate the initial girth length. The model is as follows: ,in, For the first The initial calculation length for each dimension; For the optimal cross-sectional curve The total number of point clouds on the surface; For the optimal cross-sectional curve Upper 3D point coordinates; The distance is a two-dimensional Euclidean distance; Point cloud sorting adopts polar coordinate sorting method: project the cross-sectional point cloud onto a two-dimensional plane, calculate the polar angle of each point relative to the center of the cross-section, and sort them in ascending order of polar angle to ensure the formation of closed curves.

[0050] In this embodiment, the surface correction model is used to calculate the final accurate dimensions. The model is as follows: ,in, This is a preset correction constant; To preset the standard human body curvature; For the first The final dimensions of an individual's body circumference; Curvature outlier filtering: Before calculating the correction, the system first calculates the average and standard deviation of the curvature of all points on the cross section. Points with curvature exceeding three times the average standard deviation are considered outliers and removed to avoid measurement errors caused by tattoos, scars, clothing wrinkles, etc.

[0051] Separate measurement for left and right halves: The human body is divided into left and right halves, and the optimal cross-section and dimensions for each half are calculated separately. When the deviation of the left and right shoulder width exceeds 5mm or the deviation of the left and right chest circumference exceeds 8mm, the user is prompted on the interactive interface that "left-right asymmetry has been detected, and the left and right dimensions have been measured separately." The left and right dimensions are then applied separately in the subsequent pattern generation, as shown in Table 1, which lists the parameter values ​​for different body parts: Table 1. Parameter values ​​for different parts In this embodiment, the digital interaction module includes user clothing preference settings. A "Wearing Preferences" drop-down menu is set at the top of the virtual fitting interface, providing three options: tight, standard, and loose. The default selection is standard. After the user selects, the system automatically adjusts the optimal ease: tight: the optimal ease for all measurements is reduced by 1cm; standard: the preset optimal ease is used; loose: the optimal ease for all measurements is increased by 2cm. The system will record the user's historical preferences and apply them automatically the next time they log in.

[0052] In this embodiment, the formula for calculating the fabric attribute matching factor is: ,in, The elastic modulus of the fabric; The elastic modulus of standard cotton fabric is determined using the standards for measuring breaking strength and elongation at break, with the measured values ​​being [values ​​to be filled in]. ,in, The value range is [0.5, 1.5]. When the calculation result is less than 0.5, take 0.5, and when it is greater than 1.5, take 1.5.

[0053] Optimal ease allowance for different clothing categories (Unit: cm), as shown in Table 2: Table 2 Optimal Ease Measurement Table for Different Clothing Categories Tolerance coefficient The tolerance coefficient is determined based on the degree of influence of different parts on wearing comfort. Waist circumference has the greatest impact on comfort, so it has the smallest tolerance coefficient (0.5); chest circumference has the second greatest impact on comfort, with a tolerance coefficient of 0.8; hip circumference and shoulder width have tolerance coefficients of 0.7 and 0.6, respectively.

[0054] In this embodiment, the matching score of all sample garments is calculated, and the sample garments with a score ≥ 0.6 are sorted from high to low, and the top 3 are recommended to the user. When all sample garments score <0.6, the system recommends the 3 sample garments with the highest scores and displays a message on the screen: "The current sample garment may need major adjustments. We recommend trying it on and providing detailed feedback." The recommendation results are simultaneously sent to the control unit in the second space. The control unit automatically checks the storage compartment corresponding to the highest-scoring sample garment. If the sample garment is in place, the electromagnetic electric lock is unlocked and the LED indicator turns green. Users can click the "All Styles" button to browse all styles and select any style for virtual or physical try-on.

[0055] In this embodiment, the clothing deformation mesh driving algorithm of the virtual fitting rendering module is as follows: The system uses a linear hybrid skinning (LBS) algorithm combined with position-based dynamics (PBD) to simulate clothing deformation and wrinkles. The specific steps are as follows: Skeletal binding: Pre-bind the 3D model of each garment and set 12 key skeletal points (left shoulder, right shoulder, left elbow, right elbow, waist, left hip, right hip, left knee, right knee, left ankle, right ankle, and neck). Circumference to skeleton scaling factor conversion: based on predicted clothing circumference The ratio to the base girth is used to calculate the scaling factor for the corresponding bone point; Linear blending skin: The displacement of each vertex is obtained by the weighted average of the displacements of all bones affecting that vertex, and the weight values ​​are preset when the clothing model is created; Wrinkle Simulation: The PBD algorithm is used to simulate the physical properties of clothing, taking into account the elastic coefficient, damping coefficient and gravitational acceleration of the fabric, to generate a natural wrinkle effect.

[0056] In this embodiment, the virtual fitting rendering module is used to predict the circumference after wearing the garment, and the prediction results are used to drive the garment deformation mesh. The value is 0.1cm.

[0057] Basal relaxation amount (Unit: cm) Values: Bust 2.0, Waist 1.5, Hips 2.0, Shoulder width 1.0.

[0058] The system has a built-in database containing over 500 common clothing fabrics, storing parameters such as elastic modulus, thickness, areal density, and drape coefficient for each fabric. Each sample garment corresponds to a unique fabric ID, and the system automatically retrieves the corresponding parameters based on the garment selected by the user.

[0059] In this embodiment, the virtual fitting interface provides four posture options: still, raised arm, bent over, and sitting down. Users can click to switch and view the wearing effect in different postures. Static posture: Simulates the wearing effect when the user is standing naturally; Arm raising posture: Simulates the wearing effect when the user raises both arms to 90°, focusing on showing the stretch of the cuffs and waist; Bending posture: Simulates the wearing effect when the user's upper body leans forward 30°, focusing on demonstrating the fit to the back and waist; Sitting posture: Simulates the wearing effect when the user is sitting upright, focusing on the comfort of the pants' waist and hips.

[0060] In this embodiment, the hardware configuration of the second space control unit is as follows: it uses an STM32F103ZET6 microcontroller as the main controller, is equipped with an RS485 serial communication interface to connect with the digital interaction module, and supports up to 120 relay outputs to control the electromagnetic electric latch. Control logic: Receives a list of recommended sample garment IDs sent by the digital interaction module; Check the status of the infrared presence detection sensor and RFID reader in the corresponding storage compartment to confirm that the sample garment is in place and the RFID tag is matched; If the status is normal, the electromagnetic electric latch will unlock and the LED indicator will turn green. When multiple sample garments have the same score, they will be sorted by historical order popularity, which is updated monthly by the system backend. When the user takes out the sample garment, the infrared sensor detects that the sample garment is not in place, automatically closes and locks the storage compartment, and the LED indicator turns red; When a user returns a sample garment, they place it in a storage compartment. An RFID reader reads the sample garment's tag, and after confirming that everything is correct, it automatically locks the storage compartment and updates the sample garment's fitting count.

[0061] In this embodiment, the interactive interface design of the physical try-on feedback module is as follows: The virtual try-on feedback interface on the mirrored display screen includes the following elements: Human body diagram: Marking 8 adjustable parts: chest, waist, hip, shoulder width, garment length, sleeve length, collar height, and trouser length; Adjust the slider: Each part corresponds to one slider, with a sliding range of -5cm to +5cm and a step size of 0.5cm. Positive values ​​indicate increasing and negative values ​​indicate decreasing. Tightness level selection: Each part provides five radio buttons: "Very loose, Slightly loose, Standard, Slightly tight, Very tight", which are mapped to -1.0, -0.5, 0, +0.5, and +1.0 respectively; a continuous slider is also provided, which can input any value between -1 and 1; Dynamic comfort feedback: Three dynamic comfort options have been added: cuffs when raising arms, waist when bending over, and hips when sitting down. Users can choose between too tight, fit, and too loose.

[0062] In this embodiment, the physical try-on feedback module is equipped with a physical-virtual difference quantification algorithm to calculate the secondary adjustment vector. The weights and tightness adjustment coefficients for different clothing categories are shown in Table 3. Table 3. Weighting and Tightness Adjustment Coefficients for Different Clothing Categories In this embodiment, the upper limit of the secondary adjustment amount for a single part is ±5cm. When the adjustment amount entered by the user exceeds the upper limit, the system automatically limits it to ±5cm and prompts "The adjustment amount has reached the upper limit". The system allows users to make up to 3 adjustments. If they are still not satisfied after 3 adjustments, the system will automatically pop up the contact information for customer service, and a professional tailor will provide one-on-one guidance. After the adjustment is completed, the system updates the local dimensions of the 3D digital human body model. The updated dimensions are... This triggers the virtual try-on rendering module to re-render, where the amount of secondary adjustment made by the user each time... The update size is calculated using a cumulative method, meaning the size after the nth adjustment is: When the cumulative adjustment exceeds ±5cm, the system automatically limits it to ±5cm and prompts "The cumulative adjustment has reached the upper limit. It is recommended to contact customer service".

[0063] In this embodiment, the final version parameter vector is calculated as follows: The order generation module is used to generate the final pattern parameter vector. The customized order is generated by packaging the clothing style, fabric, and color code confirmed by the user. Obtain the basic pattern parameter vector: ; Obtain the final confirmed girth vector from the user: ; Obtain fabric property correction vector Each of them According to the The fabric's elastic modulus, thickness, and drape coefficient are preset for each circumference, and the calculation formula is as follows: Therefore, it can be calculated that ; To prevent division by zero, the value is taken as... ; Dimension weight vector The weighting of different body parts on the garment's pattern is determined based on their respective influences. Bust circumference has the greatest impact, with a weight of 1.2; waist circumference and shoulder width are next, with a weight of 1.0; hip circumference has a weight of 1.1; and garment length and sleeve length have a relatively small impact, with a weight of 0.9. The value range is [0.7, 1.0]. When the calculation result is less than 0.7, it is taken as 0.7, and when it is greater than 1.0, it is taken as 1.0.

[0064] In this embodiment, the factory-side interface protocol includes the following data transmission format: It supports exporting three CAD formats: Graphical Exchange Format (DXF), Gerber, and Lectra, and also includes JSON format order information, such as order number, user ID, style ID, fabric ID, color ID, quantity, and delivery date. Automatic modification logic of CAD system: The factory parametric garment CAD system has a built-in pattern library corresponding to this system. After receiving the pattern file, it automatically reads the final pattern parameter vector Θ, corrects the basic pattern element by element, and generates production cutting paper patterns. MES System Integration: Order information is simultaneously sent to the factory's Manufacturing Execution System (MES). The MES system arranges the production plan according to the order priority, and assigns production tasks to the corresponding cutting and sewing stations. Production progress feedback: The factory's MES system synchronizes the production progress (order placed, cutting in progress, sewing in progress, quality inspection in progress, and shipment completed) to the system server in real time. Users can check the order status through a mobile APP.

[0065] In this embodiment, the order modification and cancellation functions are as follows: Within 2 hours of placing an order, users can modify the size, fabric, and color parameters of the order via the mobile app or system interface. If the factory's MES system shows an order status of "order placed" but cutting has not yet started, the user can cancel the order, and the system will automatically refund the money. If the order has already entered the cutting or sewing stage, the system will prompt the user that it cannot be cancelled and provide customer service contact information for assistance.

[0066] In this embodiment, the user session management module: User identification: Supports login via mobile phone verification code and facial recognition. Users need to register an account and fill in basic information when using it for the first time. Session isolation: The system creates an independent session space for each logged-in user to store the user's measurement data, fitting records, adjustment parameters and order information, and the data of different users is completely isolated; Queue Management: When the first space 1 is in use, subsequent users can scan the code to join the queue via mobile APP. The system displays the number of people in the queue and the estimated waiting time in real time. Data storage: The user's body measurement data and size information are permanently stored in the cloud, and the user can directly access them the next time they use the device without having to take measurements again.

[0067] In this embodiment, the sample garment lifecycle management module includes: RFID tag management: Each physical sample garment is sewn with a unique RFID tag inside, which records information such as the garment's style ID, size, warehousing time, number of times it has been tried on, and disinfection time. Disinfection reminder: When the sample garment has been tried on 20 times or more than 7 days have passed since the last disinfection, the system will automatically lock the corresponding storage compartment, the LED indicator will turn yellow, and a disinfection reminder will be sent to the administrator; Damage detection: Administrators regularly check the status of sample garments. When a sample garment shows wear, stains, or damage, it is marked as "damaged" in the system, and the system will no longer recommend that sample garment. Inventory Management: The system provides real-time statistics on the inventory quantity of each style of sample garment. When the quantity of a certain style of sample garment falls below 2 pieces, a replenishment order is automatically generated and sent to the administrator.

[0068] In this embodiment, the exception handling module handles measurement failure exceptions. The measurement timeout is set to 3 minutes. If the user's posture still does not meet the standard within 3 minutes, the system will prompt "Measurement timeout" and provide the option to manually enter the dimensions. If a camera malfunctions or insufficient light prevents the generation of a 3D model, the system will display the message "Device malfunction, please try again later" and automatically switch to 2D image estimation mode to estimate key dimensions using front and side photos uploaded by the user. If the estimation accuracy is insufficient, the system will prompt the user to contact customer service for professional measurement.

[0069] In this embodiment, hardware fault handling is performed as follows: The system monitors the operating status of hardware such as 3D cameras, electromagnetic electric locks, infrared sensors, and RFID readers in real time. When a hardware failure occurs, a clear error code and customer service contact information are displayed on the mirror screen, and the administrator is notified through the backend. Each storage compartment is equipped with a manual unlocking device, allowing administrators to open the compartment in an emergency using a key.

[0070] In this embodiment, network fault handling is performed as follows: The design allows for an offline working mode, where local functions such as body measurement, virtual fitting, physical fitting, and feedback adjustments can run offline, with data temporarily encrypted and stored on a local server. Once the network is restored, the system will automatically synchronize the locally cached data to the cloud server; If a network failure occurs when an order is generated, the system will display the message "The order has been saved and will be automatically submitted after the network is restored," and generate an order voucher for the user to save.

[0071] The integrated clothing customization method based on AI technology provided by this invention, such as... Figure 2 As shown, the specific steps are as follows: Step 1: User identification and session creation: Users log in to the system via mobile phone verification code or facial recognition; After the system verifies the user's identity, it creates an independent session and assigns a unique user ID; If the first space is in use, the system will prompt the user to queue and display the estimated waiting time.

[0072] Step 2: Body Measurement Data Collection Users enter the first space and select a black Lycra bodysuit of the appropriate size from the lockers to change into; The user stands on the circular standing marker in the center of the ground, and the guidance unit guides the user to adjust their posture through digital human animation and voice. The system uses the MediaPipe algorithm to detect the user's posture in real time. When all joint angles meet the standard and remain stable for 2 seconds, it automatically triggers 8 3D cameras to capture images simultaneously. The camera sends the captured body images from eight angles to the size extraction module.

[0073] Step 3: 3D Human Model Construction and Size Extraction: The size extraction module performs statistical filtering, radius filtering, and Poisson reconstruction on the image data to generate a human triangular mesh model. Facial blurring is applied to 3D models to protect user privacy; Candidate cross sections are generated at each circumference location, and the optimal cross section is selected using the optimal circumference cross section model; The initial girth length is calculated using the girth curve closure length model; Filter out curvature anomalies and calculate the final accurate dimensions using a surface correction model; The 3D digital human body model and key size data are sent to the digital interaction module and the virtual fitting rendering module.

[0074] Step 4: Virtual try-on and sample garment recommendations: The digital interaction module generates and displays a 3D digital human body model of the user on a mirrored display screen; Users select their clothing preference (tight / standard / loose); The system calculates the matching score of all sample garments using a matching model and recommends the top 3 sample garments with the highest scores. Users can click the "All Styles" button to browse all styles and select clothing they are interested in for a virtual try-on. The virtual fitting rendering module 11 predicts the circumference after wearing the clothing based on the clothing parameters selected by the user, and drives the clothing deformation mesh to generate a static virtual fitting effect. Users can switch between dynamic postures such as raising their arms, bending over, and sitting down to see the effect of the clothing in different postures.

[0075] Step 5: Physical sample garment automatically unlocks: After the user confirms that they want to try on a sample garment, they click the "Try On" button on the display screen. The digital interaction module sends the try-on instruction and sample garment ID to the control unit in the second space; The control unit checks the status of the infrared presence detection sensor and RFID reader in the corresponding storage compartment to confirm that the sample garment is in place and the tag matches. If the status is normal, the electromagnetic electric latch will unlock and the LED indicator will turn green. The automatic sensor door opens, and the user enters the second space to retrieve the sample garment for a physical try-on.

[0076] Step 6: Physical fitting feedback and model update: After trying on the sample garment, the user returns to the first space and stands in front of the mirrored display screen. Users input the adjustment amount and tightness level for each part through the interactive interface; The physical try-on feedback module calculates the secondary adjustment vector using a physical-virtual difference quantification algorithm; The digital interaction module updates the local dimensions of the 3D digital human body model based on the secondary adjustment vector; The virtual try-on rendering module re-renders the virtual try-on effect based on the updated human body model and displays it on the mirror screen. Users can repeat steps S5-S6 up to 3 times for adjustments.

[0077] Step 7: Order generation and production integration: After confirming the final clothing parameters, the user clicks the "Place Order" button on the display screen; The order generation module calculates the final pattern parameter vector; The final pattern parameter vector, style information, fabric information, and color information are packaged to generate a custom order; Orders are sent to both the factory's MES system and the parametric garment CAD system simultaneously. The factory's CAD system automatically reads the pattern parameters and generates production cutting patterns; the MES system arranges production plans and organizes cutting, sewing, and quality inspection. Users can check order production progress and logistics information in real time through a mobile app.

[0078] Step 8: Sample garment return and management: After the user has finished trying on the garment, they will put the sample garment into the corresponding storage compartment. The RFID reader reads the sample garment tag and automatically locks the storage compartment after confirming that it is correct. The system updates the number of times the sample garment has been tried on and the last time it was used; If the number of times the item has been tried on reaches 20 or more than 7 days have passed since the last disinfection, the system will automatically lock the storage compartment and prompt the administrator to disinfect it.

[0079] In this embodiment, to verify the technical effect of the present invention, the following three sets of comparative experiments were conducted: One hundred volunteers aged 18–60 years with a body mass index (BMI) between 18.5 and 28.0 were selected. Their body measurements—chest, waist, and hip circumference—were taken manually using both this system and a professional tailor's measuring tape. The experimental results are shown in Table 4. Table 4 Experimental Results Experimental results show that the measurement accuracy of this system is better than that of traditional manual measurement, with an average error reduction of about 68%, which can meet the millimeter-level accuracy requirements of most clothing customization scenarios.

[0080] In this embodiment, 200 users were selected and randomly divided into two groups of 100 each. The experimental group used the system's multidimensional matching model for sample garment recommendations, while the control group used the traditional height-weight + size chart method. After trying on the recommended sample garments, users rated the fit (1-5 points, ≥4 points for satisfaction). The experimental results are shown in Table 5: Table 5 Experimental Results Experimental results show that the accuracy of sample garment recommendation in this system is about 71% higher than that of traditional methods, reducing the average number of times users try on clothes and shortening the time for customized services per customer.

[0081] In this embodiment, 50 users were selected to virtually try on the same suit using both this system and a mainstream virtual fitting software. The results were then compared with the actual fit, and the fit was rated (1-5 points). The experimental results are shown in Table 6. Table 6 Experimental Results Experimental results show that the virtual fitting of this system has a significantly higher fit than existing technologies, with a similarity of 85% to the physical fitting effect, providing users with accurate fitting references.

[0082] In this embodiment, for example, custom-made men's suits: In response to the high requirements for the fit and strict control of ease in men's suits, the system has made the following parameter adjustments: Optimal ease allowance: Bust +6cm, Waist +4cm, Hips +5cm, Shoulder width +1cm; Tolerance coefficient: Bust 0.7, Waist 0.4, Hips 0.6, Shoulder width 0.5; Weight parameters: =0.8, =0.2; Dimensional weight vector: Bust 1.3, Waist 1.1, Hips 1.0, Shoulder width 1.2; Tightness adjustment range coefficient: 2mm.

[0083] In this embodiment, for example, custom-made maternity dresses: In view of the special characteristics of pregnant women's body shape and the large changes in their abdomen, the system has made the following parameter adjustments: Add abdominal circumference measurement points and generate candidate cross sections every 2mm within a 10cm range above and below the navel; Optimal ease allowance: Bust +5cm, Waist +10cm, Hips +6cm; Tolerance coefficient: Waist 1.0, Bust 0.8, Hips 0.7; Weight parameters: =0.6, =0.4; Dimensional weight vector: Waist 1.5, Bust 1.0, Hips 1.0; The virtual try-on feature provides a simulated belly bulge effect at different stages of pregnancy (4 months, 6 months, 8 months).

[0084] In this embodiment, for example, children's clothing customization: In response to the large differences in children's body size and their active nature, the system has made the following parameter adjustments: Optimal ease allowance: Bust +5cm, Waist +3cm, Hips +4cm; Tolerance coefficient: 0.9 for all dimensions; Basic relaxation allowance: Increase each circumference by 0.5cm; When taking measurements, use the quick shooting mode to shorten the shooting time to 1 second per round. The user interface features a cartoonish design and includes voice guidance functionality.

[0085] In this embodiment, for example, it is customized for wheelchair users: To accommodate the specific body type and clothing needs of wheelchair users, the system adjusts the following parameters: When taking body measurements, it supports seated posture measurement, and allows adjustment of the camera angle and standard posture requirements; Add measurement points for back curvature and hip circumference; Optimal ease allowance: back +3cm, hips +4cm, pant length -2cm; Virtual try-on provides a sitting posture simulation effect; The back and hip areas are optimized during pattern generation to improve wearing comfort.

[0086] This invention enables standardized data transfer from the consumer end to the manufacturing end. The specific process is as follows: The order generation module sends customized orders to the factory's MES system, and the MES system automatically creates production work orders and assigns unique production numbers. The MES system sends the pattern parameters to the automatic cutting machine, which reads the DXF file and automatically lays out the pattern according to the fabric width for high-precision cutting. The cut fabric is transferred to the sewing station along with the production work order. Each sewing station is equipped with an electronic Kanban board that displays the process requirements and size parameters of the work order. After sewing is completed, the garment is bound to the production order using RFID tags for quality inspection. After passing quality inspection, the MES system updates the order status to "completed" and notifies the logistics system to arrange shipment. The logistics system synchronizes the tracking number to the system server, and users can check the logistics information through a mobile app.

[0087] Through the aforementioned digital management, this invention reduces the time for pattern design and paper preparation from the traditional 1-2 days to less than 1 hour. With sufficient factory capacity and fabric inventory, production can be completed in 2-3 days.

[0088] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An AI technology-based clothing integrated customization system, characterized by, include: The first and second spaces are interconnected; The first space is equipped with a mirror display screen, a body measurement data acquisition module, and a user session management module. The body measurement data acquisition module is used to collect multi-angle body image data of users, and the user session management module is used to realize multi-user data isolation and queuing management. The second space is used to store physical sample garments of various models and styles, and to unlock the electromagnetic electric lock of the corresponding model sample garment according to the user's confirmed fitting instructions. The size extraction module is connected to the body data acquisition module and is used to construct a 3D digital human body model of the user based on the multi-angle body image data, and extract key human body size data through the optimal circumference section model, circumference curve closed length model and surface correction model. A digital interaction module, connected to the size extraction module and the mirror display screen, is used to generate and display the user's 3D digital human body model and recommend clothing parameters based on the user's clothing preferences. The virtual fitting rendering module, connected to the digital interaction module, is used to predict the circumference after wearing the garment based on the key human body size data and the clothing parameters selected by the user, combined with the physical properties of the fabric, and drive the clothing deformation mesh to generate the virtual fitting effect. The physical try-on feedback module is connected to the mirror display screen and the digital interaction module. It is used to receive feedback from users on the adjustment of the circumference and non-circumference parameters of the physical try-on sample garment, generate a secondary adjustment vector through the physical-virtual difference quantization algorithm, and synchronously update the 3D digital human body model and the virtual try-on effect. The order generation module, connected to the digital interaction module and the physical try-on feedback module, is used to generate a customized order containing structured pattern parameters based on the clothing parameters finally confirmed by the user, and to provide order modification and cancellation functions.

2. The integrated clothing customization system based on AI technology according to claim 1, characterized in that, The first space is an independent, enclosed compartment, and an automatic sensor door is installed between it and the second space; A circular standing sign is placed in the center of the ground in the first space, and depth cameras are evenly installed on the surrounding walls, divided into upper and lower layers, with LED soft lights installed on the top.

3. The integrated clothing customization system based on AI technology according to claim 1, characterized in that, The body measurement data acquisition module includes an FPGA synchronous controller and a guidance unit; The FPGA synchronization controller enables synchronous triggering of the camera; The guidance unit guides users to complete standard postures through digital human animation and voice prompts, and uses a human key point detection algorithm to detect postures in real time. When all joint angles meet the standard, it automatically triggers shooting.

4. The integrated clothing customization system based on AI technology according to claim 1, characterized in that, The dimension extraction module has a built-in curvature outlier filtering algorithm. Before calculating the surface correction amount, points with curvature exceeding three times the standard deviation of the average value are regarded as outliers and removed. At the same time, the human body is divided into two halves, left and right, and the optimal circumference cross section of the left and right halves is calculated separately. When the deviation of the left and right dimensions exceeds the threshold, the user is prompted on the interactive interface and the left and right dimensions are displayed respectively.

5. The AI-based integrated clothing customization system according to claim 1, characterized in that, The digital interaction module calculates the sample garment matching score using a matching model, which is: ; in, For the first The matching score of the sample garment stored in the second space; The total number of dimensions involved in the matching; For user number The final dimensions of each circumference; For the first Sample garment The original net dimensions of each circumference; For the first The preset optimal ease for each circumference; For the first Tolerance coefficient for each dimension; Fabric attribute matching factor; Adjustments based on user clothing preferences.

6. The AI-based integrated clothing customization system according to claim 5, characterized in that, The virtual fitting rendering module uses a linear hybrid skinning algorithm combined with position-based dynamics to simulate clothing deformation and wrinkles. The girth prediction model after wearing is as follows: ; in, Predicted circumference after wearing; The basic relaxation constant; The elastic modulus of the fabric for the selected physical sample garment. For fabric thickness, For fabric surface density, The characteristic length of the fabric stretch. This is the proportionality coefficient.

7. The AI-based integrated clothing customization system according to claim 1, characterized in that, The second space is a modular storage area with multiple expandable storage compartments. Each compartment is equipped with an electromagnetic electric lock, an infrared presence detection sensor, an RFID reader, and an LED indicator.

8. The AI-based integrated clothing customization system according to claim 6, characterized in that, The interactive interface of the physical try-on feedback module supports the adjustment of eight parameters: bust, waist, hips, shoulder width, garment length, sleeve length, collar height, and trouser length. The entity-virtual difference quantization algorithm is as follows: in, , These are preset visual weights and preset tactile weights, which are pre-set by the system according to the clothing category; This represents the secondary adjustment amount for the k-th part; Adjustments based on the amount of physical fitting input by the user; Mapping values ​​for the tightness level input by the user; This is the tightness adjustment range coefficient.

9. The AI-based integrated clothing customization system according to claim 1, characterized in that, The final version parameter vector model is: ;in, For Hadama accumulation, It is a vector of all 1s. This is the overall scaling factor. To prevent division by zero minimum constant, For the preset dimensional weight vector The basic pattern parameter vector, This is the final girth dimension vector confirmed by the user. This is a vector for correcting fabric properties. This is the parameter vector for the final version.

10. A method for integrated clothing customization based on AI technology, characterized in that, include: Collect multi-angle body image data of users; A 3D digital human body model of the user is constructed based on the multi-angle body image data, and key human body size data is extracted through the optimal circumference section model, circumference curve closed length model and surface correction model. Based on the key human body size data and the clothing parameters selected by the user, the circumference after wearing is predicted by combining the physical properties of the fabric, and the clothing deformation mesh is driven to generate a virtual fitting effect. The system receives user feedback on adjustments to the circumference and non-circumference parameters of the physical try-on sample garment, generates a secondary adjustment vector through a physical-virtual difference quantization algorithm, and synchronously updates the 3D digital human body model and the virtual try-on effect. A customized order containing structured pattern parameters is generated based on the clothing parameters finally confirmed by the user, and the order modification and cancellation functions are provided.

Images