Clothing model automatic migration and virtual fitting method

Abstract

The invention provides a clothing model automatic migration and virtual fitting method, which comprises the following steps: when a virtual fitting platform receives a target clothing fitting request, firstly extracting request features, obtaining target human body model parameters, then collecting three-dimensional topological data of a human body model, and constructing adaptive clothing model parameters according to the three-dimensional topological data; and a parameter adjusting module is integrated on the virtual fitting interface. And the adaptive clothing model parameter is updated according to the material parameter of the adjusting module, after the target clothing model is obtained, migration adaptive deformation processing is performed according to the updated parameter, and the virtual fitting effect is rendered if the adaptation is successful. The reality sense of virtual fitting and the user experience can be improved, the fitting efficiency is improved, and the operation cost of a clothing enterprise is reduced.

Description

Technical Field

[0001] This invention relates to the field of virtual try-on technology, and more specifically, to a method for automated migration of clothing models and virtual try-on. Background Technology

[0002] Against the backdrop of increasingly prominent digital and personalized consumption trends, virtual try-on technology has emerged and developed rapidly, aiming to provide consumers with a convenient, efficient, and personalized clothing try-on experience, while helping apparel companies reduce costs and improve operational efficiency. Existing virtual try-on technologies primarily utilize computer graphics and image processing techniques to demonstrate the effect of clothing on a virtual human body by constructing virtual human and clothing models. However, existing technologies have numerous problems and shortcomings in their implementation.

[0003] On the one hand, existing virtual try-on technologies have significant shortcomings in terms of clothing model adaptability. Traditional virtual try-on systems often struggle to quickly and accurately automate clothing models to suit different user body types and try-on needs. For example, when faced with users of different body types, the size adjustment of clothing models is not precise enough, resulting in a significant difference between the try-on effect and the actual wearing experience. Furthermore, for clothing with complex structures and textures, existing technologies are prone to deformation, distortion, and texture misalignment during the adaptation process, affecting the realism of the virtual try-on and the user experience.

[0004] On the other hand, existing technologies also have many limitations in simulating the dynamic effects of virtual try-on. For dynamic try-on requests, where users want to see how clothing looks during movement, existing virtual try-on systems often struggle to accurately capture and simulate the continuity and naturalness of human movements. During motion capture and analysis, issues such as lost key points and disjointed movements can easily occur, resulting in poor adaptation of the clothing model during dynamic try-on and an inability to realistically reflect the dynamic performance of clothing in actual wear.

[0005] Furthermore, existing virtual try-on technologies also have shortcomings in material representation and physical simulation. The material properties of clothing are crucial to the presentation of the try-on effect, but current technologies are often not accurate or realistic enough in simulating the physical characteristics of clothing materials, such as elasticity, flexibility, and friction. This results in a significant gap between the virtual try-on effect and the actual wearing experience, reducing users' trust and satisfaction with virtual try-on technology.

[0006] The existing technology has at least the following problems or defects: the automatic adaptation capability of clothing models is insufficient, making it difficult to meet the needs of different user body types and fitting needs; the simulation of dynamic fitting effects is not natural and accurate enough, and cannot truly reflect the performance of clothing in motion; the material representation and physical simulation are not realistic enough, affecting the realism of virtual fitting effects and user experience. Summary of the Invention

[0007] This invention provides a method for automated migration of clothing models and virtual try-on, comprising:

[0008] In response to receiving a fitting request for a target garment through a virtual fitting platform, feature extraction is performed on the fitting request to obtain the target human body model parameters corresponding to the fitting request;

[0009] Collect the three-dimensional topological data of the target human body model;

[0010] Based on the three-dimensional topology data, construct the parameters of the adapted clothing model;

[0011] A virtual fitting interface is displayed that corresponds to the parameters of the adapted clothing model, wherein the virtual fitting interface integrates a parameter adjustment module;

[0012] The parameters of the adapted clothing model are updated according to the material parameters of the corresponding parameter adjustment module.

[0013] In response to obtaining a target clothing model through the virtual fitting platform, the target clothing model is subjected to migration adaptation deformation processing based on the updated adaptation clothing model parameters to obtain a migration result;

[0014] In response to determining that the migration result represents a successful adaptation, the virtual fitting platform is controlled to render the virtual fitting effect.

[0015] Furthermore, the target human body model parameters include body type classification identifiers and model topology identifiers; and before acquiring the three-dimensional topology data of the target human body model, the method further includes:

[0016] In response to the detection of a confirmation command applied to the fitting request interface, skeletal joint data corresponding to the target human body model are collected according to the model topology identifier, wherein the fitting request interface corresponds to the fitting request and the body type classification identifier is displayed in the fitting request interface;

[0017] Determine whether the skeletal joint data meets a preset complexity condition; in response to determining that the skeletal joint data meets the preset complexity condition, assign the corresponding...

[0018] The feature extraction command for the fitting request is sent to the model analysis engine; and

[0019] The acquisition of the three-dimensional topological data of the target human body model includes:

[0020] In response to receiving the feature extraction completion response returned by the model analysis engine, the three-dimensional topological data of the target human body model is collected.

[0021] Furthermore, the step of constructing the parameters for the adapted clothing model based on the three-dimensional topology data includes:

[0022] The three-dimensional topology data is subjected to key point integrity verification processing to obtain verification results; in response to determining that the verification results represent the integrity of key points, the parameters of the adapted clothing model are constructed based on the three-dimensional topology data.

[0023] Furthermore, the fitting request is a dynamic fitting request, and the three-dimensional topology data is temporal motion capture data; and

[0024] The key point integrity verification process for the three-dimensional topology data, to obtain the verification result, includes:

[0025] The motion continuity analysis of the time-series motion capture data is performed to obtain the motion continuity results as the verification results.

[0026] Furthermore, the step of constructing the parameters for the adapted clothing model based on the three-dimensional topology data includes:

[0027] The temporal motion capture data is subjected to motion decomposition operations to obtain a sequence of motion frames; for each motion frame in the sequence of motion frames, the following steps are performed:

[0028] Determine whether a complete skeletal joint exists in the action frame;

[0029] In response to determining that there are complete skeletal joints in the action frame, the coordinates of the joints in the action frame are normalized to obtain a standard joint set.

[0030] Motion amplitude analysis is performed on the standard joint set to obtain joint motion amplitude parameters;

[0031] Stability analysis was performed on the standard joint set to obtain joint stability parameters;

[0032] Collision detection calculations are performed on the standard set of joints to obtain joint collision parameters;

[0033] Based on the joint motion amplitude parameters, the joint stability parameters, and the joint collision parameters, construct the motion frame parameters;

[0034] Based on the obtained parameters of each action frame, select the target set of joints that meet the preset action conditions from the obtained set of standard joints.

[0035] Based on the joint motion amplitude parameters, joint stability parameters and joint collision parameters corresponding to the target joint set, motion compensation calculation is performed on the target joint set to obtain the compensated joint set.

[0036] The compensated joint set is subjected to feature encoding to obtain motion feature vectors; based on the motion feature vectors and the body type classification identifier, the parameters of the adapted clothing model are constructed.

[0037] Further, the step of performing motion amplitude analysis on the standard joint set to obtain joint motion amplitude parameters includes:

[0038] The velocity vector of the standard joint set is calculated to obtain the joint velocity matrix; the joints in the standard joint set that satisfy the preset static threshold are removed to obtain the dynamic joint set.

[0039] For each joint in the dynamic joint set, the following steps are performed: the joints adjacent to the joint in the joint velocity matrix are determined as adjacent joints;

[0040] A first displacement difference and a second displacement difference are generated based on the displacement vectors of the adjacent joints and the displacement vectors of the joints.

[0041] The sum of the magnitudes of the obtained first displacement differences and the second displacement differences is determined as the third displacement difference;

[0042] The number of each joint in the dynamic joint set is determined as the total number of joints.

[0043] The ratio of the third displacement difference to the total number of joints is determined as the joint motion amplitude parameter corresponding to the action frame.

[0044] Further, the step of performing motion compensation calculations on the target joint set based on the joint motion amplitude parameters, joint stability parameters, and joint collision parameters corresponding to the target joint set to obtain a compensated joint set includes:

[0045] In response to determining that the joint motion amplitude parameters corresponding to the target joint set meet the amplitude compensation condition, the target joint set is subjected to motion smoothing deformation to update the target joint set;

[0046] In response to determining that the joint stability parameters corresponding to the target joint set meet the stability compensation condition, the target joint set is subjected to trajectory optimization deformation to update the target joint set;

[0047] In response to determining that the collision parameters of the target joint set satisfy the collision avoidance conditions, the target joint set is spatially offset and deformed to update the target joint set.

[0048] Perform a topology consistency check operation on the updated target key set to update the target key set;

[0049] The updated target set of joints is determined as the compensated set of joints.

[0050] Furthermore, the fitting request is a static fitting request, and the three-dimensional topology data is static body shape scan data; and the key point integrity verification processing of the three-dimensional topology data to obtain the verification result includes:

[0051] Perform body topology verification processing on the static body shape scan data, and obtain the topology verification result as the verification result; and

[0052] The step of constructing the parameters for the adapted clothing model based on the three-dimensional topology data includes:

[0053] The static body shape scan data is subjected to noise point filtering operation to obtain the filtered point cloud;

[0054] The curvature distribution of the filtered point cloud is calculated to obtain a curvature distribution map;

[0055] The curvature distribution map is partitioned to obtain a body shape partition sequence;

[0056] For each partition in the body size partition sequence, perform the following steps:

[0057] Determine the curvature variance value corresponding to the partition;

[0058] Determine the surface area percentage corresponding to the partition;

[0059] In response to determining that the curvature variance value and the surface area ratio meet preset abnormal conditions, the partition is removed from the body shape partition sequence to update the body shape partition sequence;

[0060] Feature extraction is performed on each partition in the updated body shape partition sequence to obtain a body shape feature vector;

[0061] For any two obtained body shape feature vectors, determine the similarity parameter between the two body shape feature vectors;

[0062] Select the similarity parameter that meets the preset similarity conditions from the determined similarity parameters as the target similarity parameter;

[0063] The two body shape feature vectors corresponding to the target similarity parameter and the body shape classification identifier are used to construct the parameters of the clothing adaptation model.

[0064] Furthermore, after controlling the virtual fitting platform to render the virtual fitting effect, the method further includes:

[0065] Extract the set of material physical properties corresponding to the target clothing model from the clothing knowledge graph;

[0066] Construct a constraint matrix for fabric dynamics simulation based on the set of physical properties of the material;

[0067] Based on the fabric dynamics simulation constraint matrix, the virtual try-on effect is physically simulated and rendered to generate a physically simulated try-on effect.

[0068] Further, the step of performing transfer adaptation deformation processing on the target clothing model based on the updated adapted clothing model parameters to obtain the transfer result includes:

[0069] The security verification engine is invoked to perform digital signature verification calculations on the target clothing model.

[0070] In response to successful verification, based on the motion feature vector in the parameters of the adapted clothing model...

[0071] Using the body shape feature vector, calculate the vertex deformation weight matrix of the target clothing model;

[0072] The migration result is generated based on the vertex deformation weight matrix.

[0073] The embodiments of the present invention have at least the following beneficial effects:

[0074] 1. By automatically extracting the features of virtual try-on requests and constructing parameters for the adapted clothing model, this invention can quickly generate accurately adapted clothing models for different users' body models. This effectively solves the problem in existing technologies where clothing models are difficult to adapt quickly and accurately to user body shapes, significantly improving the personalized experience and adaptation efficiency of virtual try-on. It enables users to obtain a more fitting try-on effect for their own body shape during the virtual try-on process, enhancing user satisfaction and trust in virtual try-on technology.

[0075] 2. This invention addresses dynamic try-on requests by performing a series of processes on temporal motion capture data, including key point integrity verification, motion amplitude analysis, stability analysis, and collision detection. This process constructs parameters to adapt to the clothing model, achieving accurate simulation of the dynamic try-on effect. This effectively solves the problems of disjointed movements and lost key points in existing dynamic try-on simulation technologies, making the try-on effect of clothing during movement more natural and realistic. Users can clearly see the dynamic performance of clothing under different movements, thus better evaluating the wearing effect and comfort of the clothing, further enhancing the realism and user experience of virtual try-on.

[0076] 3. After rendering the virtual try-on effect, this invention extracts the set of material physical properties corresponding to the target clothing model from the clothing knowledge graph, constructs a fabric dynamics simulation constraint matrix, and performs physical simulation rendering based on this matrix to generate a physically simulated try-on effect. This innovative approach effectively solves the problem of insufficient realism in material representation and physical simulation in existing technologies, making the virtual try-on effect closer to the real wearing experience in terms of material texture and physical properties. It provides users with a more realistic and reliable virtual try-on experience, while also providing clothing companies with more accurate product displays and design references, helping to improve their market competitiveness and operational efficiency. Attached Figure Description

[0077] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein:

[0078] Figure 1 This is a flowchart illustrating an embodiment of the automated migration and virtual try-on method for clothing models provided by the present invention. Detailed Implementation

[0079] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following drawings indicate similar items; therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0080] This application proposes an automated migration and virtual try-on method for clothing models, comprising: S1, in response to receiving a try-on request for a target garment through a virtual try-on platform, extracting features from the try-on request to obtain target human body model parameters corresponding to the try-on request; S2, acquiring three-dimensional topological data of the target human body model; S3, constructing adapted clothing model parameters based on the three-dimensional topological data; S4, displaying a virtual try-on interface corresponding to the adapted clothing model parameters, wherein the virtual try-on interface integrates a parameter adjustment module; S5, updating the adapted clothing model parameters according to the configured material parameters corresponding to the parameter adjustment module; S6, in response to obtaining a target clothing model through the virtual try-on platform, performing migration, adaptation, and deformation processing on the target clothing model according to the updated adapted clothing model parameters to obtain a migration result; S7, in response to determining that the migration result indicates successful adaptation, controlling the virtual try-on platform to render the virtual try-on effect.

[0081] The target human body model parameters refer to the set of user body shape feature data obtained by analyzing fitting requests. This can be achieved using natural language processing technology or a pre-defined template matching algorithm to accurately capture differences in user body shape features and provide benchmark parameters for subsequent clothing model adaptation. 3D topology data refers to 3D point cloud or mesh data containing the geometric shape and spatial structure relationship of the human body surface. This can be achieved using laser scanning or depth camera acquisition technology to establish an accurate digital model of the human body shape. Adapted clothing model parameters refer to the set of clothing adjustment parameters dynamically generated based on human body shape features. This can be achieved using machine learning models or physical simulation algorithms to guide the automated deformation adaptation of the clothing model. The virtual fitting interface parameter adjustment module refers to an interactive graphical interface component with real-time parameter adjustment capabilities. This can be developed using visual programming tools or a UI framework to allow users to personalize adjustments to clothing materials. Migration adaptation deformation processing refers to the geometric transformation process of adapting a standard clothing model to a specific human body model. This can be achieved using skeletal binding technology or vertex deformation algorithms to eliminate geometric differences between the clothing model and the human body model. Virtual try-on rendering control refers to the process of generating instructions for graphics rendering of the adapted 3D scene. Specifically, it can be implemented using GPU-accelerated rendering pipelines or ray tracing technology to generate virtual try-on visualization results with realistic lighting and shadow effects.

[0082] The core innovation of this application lies in the optimization of the entire process of virtual try-on, including human feature capture, clothing parameter adaptation, dynamic effect simulation, and physical property rendering, through the construction of a multi-dimensional data collaborative optimization mechanism and a closed-loop verification system. This solution solves the problem of inaccurate body feature recognition in traditional technologies by combining feature extraction and 3D topology analysis; overcomes the distortion of dynamic effects caused by static model adaptation through a dynamic coupling mechanism of parameterized adjustment and transfer adaptation; and finally, ensures a dual improvement in both geometric adaptation accuracy and physical realism in the virtual try-on effect through closed-loop control of physical simulation rendering and effect verification.

[0083] The working process and principle of this application are as follows: the virtual fitting platform receives a fitting request for the target garment, extracts features from the fitting request, and obtains the parameters of the target human body model. This step establishes the basic data framework for personalized adaptation and solves the problem of inaccurate capture of user body features.

[0084] Next, 3D topological data of the target human body model is collected, and parameters for the adapted clothing model are constructed based on this data. This transforms human morphological features into quantifiable criteria for adjusting the clothing model, enhancing the automation of model adaptation.

[0085] The system displays a virtual try-on interface with parameters corresponding to the adapted clothing model, which integrates a parameter adjustment module. This design allows users to adjust material parameters in real time, compensating for the lack of flexibility in adjusting physical properties.

[0086] Based on the configured material parameters, the parameters of the adapted clothing model are updated. This step enables real-time matching between the clothing model and human movements or body shapes, overcoming the problem of dynamic effect distortion caused by static adaptation.

[0087] Once the virtual fitting platform obtains the target clothing model, the system performs migration adaptation deformation processing on the target clothing model based on the updated adaptation clothing model parameters to obtain the migration result.

[0088] Finally, after confirming that the migration result representation is successfully adapted, the virtual try-on platform is controlled to render the virtual try-on effect. This ensures the realism of the virtual try-on effect at the physical simulation level and the integrity of the visual presentation, forming a closed-loop optimization mechanism.

[0089] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0090] The virtual fitting platform receives user requests to try on desired clothing. The system extracts features from these requests to obtain parameters of the target human body model, including body type classification identifiers and model topology identifiers.

[0091] The acquisition device obtains 3D topological data of the target human body model. For dynamic fitting requests, it acquires temporal motion capture data; for static fitting requests, it acquires static body shape scan data.

[0092] The system constructs adaptive clothing model parameters based on 3D topology data. For dynamic fitting requests, it performs motion decomposition calculations to obtain a sequence of motion frames. For each motion frame, it performs joint coordinate normalization, motion amplitude analysis, stability analysis, and collision detection. For static fitting requests, it performs noise point filtering, curvature distribution calculation, and body shape partitioning.

[0093] The virtual try-on interface displays the parameters of the adapted clothing model and integrates a parameter adjustment module. Users can use this module to adjust material parameters, such as reflectivity and elasticity coefficient.

[0094] The system updates the parameters of the adapted clothing model based on the material parameters configured by the user. This includes motion compensation for dynamic try-ons and updating the body shape feature vectors for static try-ons.

[0095] After acquiring the target clothing model, the virtual fitting platform performs migration and adaptation deformation processing based on the updated parameters of the clothing model. This involves calculating the vertex deformation weight matrix and executing the mesh deformation algorithm.

[0096] After the system verifies the migration results and confirms successful adaptation, it controls the rendering engine to generate a virtual try-on effect. The rendering process considers fabric dynamics and lighting models to present a realistic clothing effect.

[0097] This application further proposes that the target human body model parameters include body type classification identifiers and model topology identifiers; before collecting the three-dimensional topology data of the target human body model, the method further includes: in response to detecting a confirmation instruction acting on the fitting request interface, collecting skeletal joint data of the corresponding target human body model according to the model topology identifier, wherein the fitting request interface corresponds to the fitting request and displays the body type classification identifier; determining whether the skeletal joint data meets the preset complexity condition; in response to determining that the skeletal joint data meets the preset complexity condition, sending the feature extraction instruction of the corresponding fitting request to the model analysis engine; collecting the three-dimensional topology data of the target human body model, including: in response to receiving the feature extraction completion response returned by the model analysis engine, collecting the three-dimensional topology data of the target human body model.

[0098] The model topology identifier can be configured as a dataset containing skeletal hierarchical structure information or joint connection rules. For example, the skeletal hierarchy index table in the topology identifier can define the joint acquisition order. Preset complexity conditions can be set as a threshold for the number of joints, a spatial distribution density threshold, or a motion trajectory sampling rate. For example, in a dynamic fitting scenario, at least 30 valid joints are required, and the distance between adjacent joints must not exceed 5 cm. The sending of feature extraction commands and the response of the model analysis engine are decoupled through an asynchronous communication protocol, such as using a message queue mechanism to ensure the reliability of command transmission. The acquisition of 3D topology data is constrained to occur after the feature extraction is completed and a response is triggered, such as by starting the data acquisition module through a hardware interrupt signal or a software callback function.

[0099] Specifically, when the try-on request interface receives a confirmation command triggered by the user, the system initiates joint point data acquisition based on the predefined skeletal structure rules in the model topology identifier. For example, if the model topology identifier is marked as a dynamic try-on mode, the system prioritizes acquiring the temporal joint point coordinate sequence returned by the motion capture device. During the acquisition process, if the number of joint points does not reach a preset threshold or the spatial distribution density is lower than the standard value, it is determined that the complexity condition is not met and the subsequent process is terminated. After the data passes verification, the feature extraction command is pushed to the message queue of the model analysis engine, triggering the feature extraction thread to start. The response signal returned by the model analysis engine after completing feature extraction activates the 3D topology data acquisition module through an event listening mechanism. This closed-loop control mechanism ensures that the timing of data acquisition and feature extraction strictly corresponds. For example, in a dynamic try-on scenario, after the feature extraction response signal is triggered, the 3D motion capture device immediately starts high-precision data acquisition, avoiding mismatch between motion frames and topology data due to timing misalignment. Thus, the pre-processing for quality verification of skeletal joint point data effectively prevents invalid data from entering the subsequent modeling process, while realizing on-demand allocation of computing resources through command distribution and response mechanisms.

[0100] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0101] The virtual fitting platform's user interface includes a fitting request interface with a body type classification label display area and a confirmation button. After selecting a body type classification label, the user clicks the confirmation button to trigger a confirmation command. The system responds to the confirmation command by collecting skeletal joint data of the target human body model based on pre-defined model topology labels.

[0102] Skeletal joint data acquisition can be accomplished using depth cameras or motion capture devices. The system verifies the complexity of the acquired skeletal joint data, with verification criteria including the number of joints, spatial distribution, and motion continuity. For example, it requires at least 15 joints, including key locations such as the head, neck, shoulders, elbows, wrists, torso, hips, knees, and ankles; spatial distribution requires the joints to form a complete human skeletal structure in three-dimensional space; and motion continuity is evaluated by calculating the displacement of corresponding joints in adjacent time frames.

[0103] When the skeletal joint data meets the preset complexity condition, the system sends a feature extraction command to the model analysis engine. The model analysis engine executes the feature extraction algorithm, which may include steps such as skeletal structure analysis and body shape feature extraction. After feature extraction is complete, the model analysis engine returns a feature extraction completion response.

[0104] After receiving the feature extraction completion response, the system initiates the 3D topology data acquisition program. This program may use a 3D scanner or multi-view camera array to acquire 3D point cloud data of the user's entire body. During the acquisition process, the system will guide the user to maintain a specific posture to ensure data quality. The acquired 3D topology data will be used in the subsequent clothing model adaptation process.

[0105] This application further proposes to perform key point integrity verification processing on 3D topology data to obtain verification results; in response to determining that the verification results represent the integrity of key points, the parameters of the adapted clothing model are constructed based on the 3D topology data.

[0106] The keypoint integrity verification process includes verifying the existence and topological relationships of preset keypoints in the 3D topology data. For example, in a dynamic fitting scene, the keypoint integrity verification process can perform continuity analysis on the displacement differences of joints between adjacent frames in the temporal motion capture data. If the displacement difference exceeds a preset threshold, it is determined that a keypoint is missing. In a static fitting scene, the point cloud density of key areas in the body shape scanning data can be statistically analyzed. When the density is lower than a preset threshold, the topological structure is determined to be incomplete. The verification results are represented by binary identifiers or integrity scores. When the score is higher than a preset threshold, the parameter construction process is triggered.

[0107] The key point integrity verification process, together with the prior skeletal joint data acquisition, forms a dual data quality verification mechanism. After the skeletal joint data meets the preset complexity conditions, the acquisition and integrity verification of the 3D topology data further ensures that the input data meets the parameter construction requirements. For example, after the number of skeletal joints reaches the preset value, the 3D topology data also needs to verify whether the spatial distribution of joints in key areas such as the shoulder and waist conforms to human kinematic constraints, thus eliminating invalid data input through dual verification.

[0108] Specifically, in dynamic fitting request scenarios, after temporal motion capture data is decomposed into a sequence of motion frames, keypoint integrity verification is performed on each frame. For example, for running motion capture data, the existence of the coordinates of the hip and knee joints in each frame is verified, and the displacement changes are verified to conform to biomechanical laws. When the displacement of the same joint point in five consecutive frames changes by more than 50 millimeters, the keypoint data is judged to be abnormal, and parameter construction for the current frame is stopped. Only after all motion frames have passed verification are the parameters of the adapted clothing model, including motion amplitude and stability parameters, generated based on the complete data, thereby avoiding distortion of the overall motion simulation due to abnormal data in a single frame.

[0109] In static virtual try-on scenarios, after noise filtering, the body shape scan data undergoes topological verification on key body shape regions. For example, for the chest and waist areas, the integrity of the point cloud closed loops and the continuity of curvature distribution are verified. When more than 10% of the point cloud in the chest area is missing, the key points are deemed incomplete, triggering a data resampling mechanism. By embedding integrity verification into parameters to construct a pre-process, it is ensured that subsequent body shape feature vector extraction is based on a complete topological structure, making the generated adaptation parameters accurately reflect human body size proportions and improving the fit of clothing model deformation in virtual try-on.

[0110] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0111] When constructing parameters for the adapted clothing model, the first step is to perform keypoint integrity verification on the 3D topology data. Specifically, after receiving the 3D topology data, the system checks the keypoints in the data using a preset verification algorithm. This algorithm may include checking whether the number of keypoints meets expectations, whether the relative positional relationships between keypoints are correct, and whether the coordinate values ​​of the keypoints are within a reasonable range.

[0112] After verification, the system obtains a verification result. This result can be a Boolean value indicating whether the keypoints are complete, or a data structure containing information about missing or abnormal keypoints. The system then evaluates the verification result. If the verification result indicates that the keypoints are complete, meaning all necessary keypoints exist and are valid, the system continues with the subsequent parameter construction steps.

[0113] After confirming the integrity of key points, the system constructs the parameters of the adapted clothing model based on the 3D topology data. This process may involve multiple steps, such as extracting key size information, calculating body shape features, and analyzing posture data. The constructed parameters of the adapted clothing model may include, but are not limited to, body size parameters such as height, weight, chest circumference, waist circumference, and hip circumference, as well as information such as body type classification and posture features.

[0114] These parameters are then used in the subsequent clothing model adaptation process to ensure that the generated virtual clothing can accurately fit the user's body characteristics and posture.

[0115] Through the above technical solution, this application can effectively improve the accuracy and reliability of the parameters for adapting clothing models. By verifying the integrity of key points before parameter construction, the system can promptly detect and eliminate incomplete or abnormal 3D topological data, avoiding parameter construction errors caused by data defects. This not only improves the accuracy of the subsequent virtual try-on process but also enhances the stability and reliability of the entire system.

[0116] Furthermore, this method improves system efficiency. By filtering out valid data at an early stage, it avoids processing invalid data during the subsequent complex parameter construction and clothing fitting processes, thus saving computational resources and time. This is especially important for virtual fitting systems that require real-time responses, significantly enhancing the user experience.

[0117] Meanwhile, the flexibility of this method is also noteworthy. Key point integrity verification can be adjusted according to different application scenarios and needs. For example, different key point verification standards can be set for different types of clothing or different body characteristics, thereby achieving more refined data quality control.

[0118] In some of the solutions described above in this application, traditional static keypoint verification methods cannot effectively capture the continuous motion features of joints in motion sequences, leading to problems such as motion breaks or data gaps in the verification of keypoint integrity in motion capture data. To address this, this application further proposes using dynamic fitting requests and 3D topology data as temporal motion capture data; and performing motion continuity analysis on the temporal motion capture data to obtain motion continuity results as verification results. The motion continuity analysis can be implemented based on trajectory correlation calculations within a time window. For example, a sliding window algorithm is used to perform differential calculations on the joint displacement vectors of adjacent frames, and the continuity of the motion trajectory is judged by setting a displacement change threshold. Furthermore, the expected motion path of the joints can be predicted using Kalman filtering or a Hidden Markov Model, and the deviation between the actual captured data and the predicted path is used as a continuity evaluation index. When the deviation exceeds a preset range, it is marked as a motion break region and a data compensation mechanism is triggered. In the motion decomposition calculation stage, the generation of the motion frame sequence needs to be correlated with the motion continuity results. For example, if the joint displacement vector of a certain motion frame deviates from the predicted path of the previous frame by more than a threshold, interpolation correction is performed on that frame or trajectory smoothing processing from adjacent frames is introduced. By dynamically adjusting the time window length and displacement change rate parameters, the continuity analysis requirements under different motion speeds can be adapted. Specifically, motion continuity analysis processing extracts acceleration, angular velocity, and displacement direction parameters from time-series data to construct multi-dimensional motion feature vectors. Based on the similarity calculation of feature vectors, motion abrupt change points between adjacent frames are identified. For example, a cosine similarity algorithm is used to analyze the directional consistency of joint velocity vectors in adjacent frames; if the similarity is lower than a preset threshold, it is determined to be a potential break area. Furthermore, by introducing a timestamp alignment mechanism, it is ensured that the sampling frequency of motion capture data matches the frame rate of dynamic fitting requests, avoiding motion distortion caused by time axis misalignment. In the data compensation mechanism, cubic spline interpolation or inverse kinematics simulation based on a physics engine can be used to reconstruct the coordinates of missing joints for break areas. For example, when an abnormal displacement of a certain joint is detected in three consecutive frames, transition data for the intermediate frame is generated based on the motion trend of the preceding and following frames. By combining the bone length ratio parameters in the body type classification identifier, the physical rationality of the interpolation results can be constrained, avoiding joint over-limit or reverse bending phenomena. Therefore, the motion continuity results not only verify the spatial distribution integrity of key points but also ensure the physical rationality of clothing model migration and adaptation during dynamic fitting by analyzing motion patterns in the time dimension. For example, in the captured data of running motion, by detecting the coordinated motion characteristics of the hip and knee joints within the stride cycle, the phase deviation of leg motion caused by sensor jitter can be identified and corrected, thereby improving the deformation accuracy of the clothing model in dynamic fitting.

[0119] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0120] The fitting request is a dynamic fitting request, and the 3D topology data is temporal motion capture data. Motion continuity analysis is performed on the temporal motion capture data, and the resulting motion continuity is used as the verification result.

[0121] Specifically, temporal motion capture data can include multiple time frames, each containing the three-dimensional coordinates of the human skeletal joints. Motion continuity analysis can be performed through the following steps:

[0122] First, the joint displacements between adjacent time frames are calculated to obtain the joint displacement vectors.

[0123] Secondly, a smoothness analysis is performed on the joint displacement vector to calculate the rate of change of the displacement vector.

[0124] Next, based on the preset continuity threshold, it is determined whether the rate of change of the displacement vector exceeds the threshold range.

[0125] Finally, the number of joints and time frames exceeding the threshold range are counted to generate motion continuity results.

[0126] Motion continuity results can include a continuity score and anomaly frame markers. The continuity score reflects the overall smoothness of the motion, while the anomaly frame markers indicate time frames where there may be motion breaks or missing data.

[0127] This application further proposes to perform motion decomposition operations on temporal motion capture data to obtain a motion frame sequence. For each motion frame in the motion frame sequence, the following steps are performed: determining whether there are complete skeletal joints in the motion frame; in response to determining that there are complete skeletal joints in the motion frame, normalizing the coordinates of the joints in the motion frame to obtain a standard joint set; performing motion amplitude analysis on the standard joint set to obtain joint motion amplitude parameters; performing stability analysis on the standard joint set to obtain joint stability parameters; performing collision detection on the standard joint set to obtain joint collision parameters; constructing motion frame parameters based on the joint motion amplitude parameters, joint stability parameters, and joint collision parameters; selecting a target joint set that meets preset motion conditions from the obtained standard joint set based on the obtained motion frame parameters; performing motion compensation operations on the target joint set based on the joint motion amplitude parameters, joint stability parameters, and joint collision parameters corresponding to the target joint set to obtain a compensated joint set; performing feature encoding processing on the compensated joint set to obtain a motion feature vector; and constructing adaptive clothing model parameters based on the motion feature vector and body type classification identifier.

[0128] The motion decomposition process employs a time window segmentation algorithm to break down continuous motion data into discrete frame sequences, for example, segmenting at a sampling rate of 30 frames per second. Normalization maps joint coordinates from different acquisition device coordinate systems to a unified reference coordinate system using a coordinate transformation matrix; for example, setting the hip joint as the origin and establishing a standard three-dimensional coordinate system. Motion amplitude analysis calculates the mean magnitude of joint displacement vectors between adjacent frames, for example, quantifying displacement changes using the Euclidean distance formula. Stability analysis calculates the second derivative of the joint trajectory using a sliding window, for example, judging trajectory smoothness using an acceleration threshold. Collision detection uses a bounding box intersection detection algorithm to identify the minimum distance between joints, for example, setting a collision threshold range of 5 mm to 10 mm.

[0129] Preset motion conditions can be set such that the motion amplitude parameter does not exceed the upper threshold, the stability parameter is higher than the lower threshold, and the collision parameter is zero. For example, the amplitude threshold can be set to 50 centimeters per second, and the stability lower limit can be set to 0.8. In motion compensation calculations, amplitude compensation uses a low-pass filter to smooth abrupt velocity changes, trajectory optimization uses Bézier curve fitting to eliminate high-frequency noise, and spatial offset is adjusted by vector projection to change joint positions. Feature encoding processing uses an autoencoder to reduce the dimensionality of joint coordinates to a 128-dimensional vector, for example, by using a three-layer fully connected neural network for compression.

[0130] Specifically, motion decomposition divides continuous motion data into independent analysis units, eliminating the masking effect of overall motion continuity on local abnormal frames. Frame-by-frame verification of skeletal joint integrity eliminates invalid frames with missing or incorrectly acquired data. Normalization eliminates data bias caused by device differences, ensuring subsequent analysis is based on a unified benchmark. Motion amplitude parameters identify clothing deformation requirements caused by vigorous movements, stability parameters avoid model parameter fluctuations caused by jitter noise, and collision parameters proactively avoid clipping problems caused by self-intersection.

[0131] Action frame parameters are constructed based on multi-dimensional parameters to comprehensively evaluate the impact weight of single-frame actions on clothing adaptation. Preset action conditions are used to select high-quality target frames and eliminate abnormal data interference. Motion compensation calculations are performed to specifically correct defects in different parameters, such as smoothing filters for frames with abrupt amplitude changes, trajectory optimization for jittery frames, and spatial offsetting for collision frames. After compensation, the keypoint set meets the requirements of physical rationality and motion continuity, providing high-quality input data for feature encoding.

[0132] Feature encoding extracts low-dimensional motion feature vectors, preserving core dynamic characteristics while reducing computational complexity. Combined with body type classification labels, adaptive clothing model parameters are constructed, enabling the clothing model to respond to dynamic motion changes and adapt to different body type characteristics. For example, in running motion, the feature vector captures the large-amplitude movement characteristics of the lower limb joints, driving the clothing model to enhance elastic deformation parameters in the hip and knee regions; in frames with non-zero collision parameters, the distance between the shoulder and elbow models is adjusted to avoid penetration. Thus, clothing deformation and human movement are precisely synchronized during dynamic try-on, significantly improving adaptation accuracy and realism.

[0133] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0134] Motion decomposition is performed on the temporal motion capture data to obtain a sequence of motion frames. For example, a sliding window method can be used to segment continuous motion data, with each window corresponding to one motion frame.

[0135] For each action frame in the action frame sequence, perform the following steps:

[0136] First, determine if complete skeletal joints exist in the motion frame. Completeness can be determined by checking if key skeletal points (such as the head, torso, and limbs) are present.

[0137] If a complete set of skeletal joints is determined to exist in the action frame, the coordinates of the joints in the action frame are normalized to obtain a standard set of joints. Normalization can be performed using the minimum-maximum normalization method, mapping the coordinate values ​​to the interval [0,1].

[0138] Motion amplitude analysis is performed on a standard set of joints to obtain joint motion amplitude parameters. The displacement distance of joints between adjacent frames can be calculated, and the average value is taken as the motion amplitude parameter.

[0139] Stability analysis is performed on a standard set of joints to obtain joint stability parameters. The smoothness of joint trajectories across multiple consecutive frames can be calculated, and stability is represented by curve fitting error.

[0140] Collision detection is performed on a standard set of joints to obtain joint collision parameters. The bounding box algorithm can be used to detect spatial interference between joints and record the number of collisions.

[0141] Action frame parameters are constructed based on joint motion amplitude parameters, joint stability parameters, and joint collision parameters. These three types of parameters can be combined into a vector to represent the action frame features.

[0142] Based on the obtained parameters of each action frame, a set of target joints that meet the preset action conditions is selected from the obtained set of standard joints. The preset conditions can be a combination of thresholds for motion amplitude, stability, and collision parameters.

[0143] Based on the joint motion amplitude parameters, joint stability parameters, and joint collision parameters corresponding to the target joint set, motion compensation calculations are performed on the target joint set to obtain the compensated joint set. A Kalman filter algorithm can be used to smooth the joint trajectories.

[0144] The compensated joint set is subjected to feature encoding to obtain action feature vectors. Principal component analysis (PCA) can be used to reduce dimensionality and extract key features.

[0145] Based on motion feature vectors and body type classification labels, parameters for an adapted clothing model are constructed. The motion feature vectors can be combined with a pre-trained body type-clothing mapping model to generate deformation parameters for the adapted clothing model.

[0146] This application further proposes to calculate the velocity matrix of a standard joint set by performing velocity vector calculation, remove joints that meet a preset static threshold to obtain a dynamic joint set, and for each joint in the dynamic joint set, determine the adjacent joints and calculate the displacement vector difference, and determine the motion amplitude parameter by the ratio of the sum of the magnitudes to the total number of joints.

[0147] The velocity vector calculation can be achieved using either a difference method or an integration method. For example, the instantaneous velocity can be obtained by dividing the coordinate change between two adjacent frames by the time interval. A preset static threshold can be set to 0.1 m / s to filter joints with displacement changes below this threshold. Adjacent joints can be determined based on direct connections in the skeletal topology, such as the shoulder and elbow joints being adjacent nodes. The displacement vector difference is generated through vector subtraction. The first displacement difference is the current node's displacement vector minus the adjacent node's displacement vector, and the second displacement difference is the adjacent node's displacement vector minus the current node's displacement vector. The sum of their magnitudes represents the degree of difference in motion direction. The introduction of the total number of joints normalizes the third displacement difference, avoiding inconsistencies in parameter scales due to different numbers of dynamic nodes.

[0148] Specifically, in the motion frame processing, dynamic joints are first identified based on velocity vector calculations, for example, by comparing each element in the velocity matrix with a static threshold. Then, for each dynamic joint, its set of adjacent nodes is traversed; for example, the hip joint corresponds to the connection between the thigh and torso, calculating the difference in their displacement vectors and accumulating the magnitude. Finally, the accumulated result is divided by the total number of dynamic nodes to generate standardized motion amplitude parameters. This process quantifies the intensity of joint coordination through local motion correlation analysis; for example, the displacement difference between the shoulder and elbow joints during arm swing reflects motion coordination. Normalization ensures the comparability of parameters between different motion frames; for example, the difference in motion amplitude between a rapid hand wave and slow walking can be accurately represented by a unified dimension, thus providing precise input for the dynamic adaptation of clothing models.

[0149] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0150] The velocity vectors of a standard set of joints are calculated to obtain the joint velocity matrix. Specifically, the velocity vector of each joint can be calculated by utilizing the changes in joint position between two adjacent frames, thereby constructing the joint velocity matrix.

[0151] The set of standard joints is obtained by removing all joints that meet a preset static threshold from the set of standard joints. For example, a speed threshold can be set, and joints with speeds below the threshold can be considered static joints and removed, while joints with speeds above the threshold are retained as the set of dynamic joints.

[0152] For each joint in the dynamic joint set, perform the following steps:

[0153] First, the joints adjacent to the given joint in the joint velocity matrix are identified as adjacent joints. Adjacent joints can be predefined based on the human skeletal structure.

[0154] Secondly, based on the displacement vectors of adjacent joints and the displacement vector of the current joint, a first displacement difference and a second displacement difference are generated. Specifically, these differences can be obtained by calculating the displacement differences between adjacent joints and the current joint in the x, y, and z directions.

[0155] Then, the sum of the magnitudes of the obtained first displacement differences and second displacement differences is determined as the third displacement difference. This step integrates the motion information of multiple adjacent joints.

[0156] Next, the number of each joint in the dynamic joint set is determined as the total number of joints.

[0157] Finally, the ratio of the third displacement difference to the total number of joints is determined as the joint motion amplitude parameter for the corresponding action frame. This step achieves the normalization of motion amplitude.

[0158] This application further proposes the following steps: in response to determining that the motion amplitude parameters of the target joint set satisfy the amplitude compensation condition, the target joint set is updated by performing motion smoothing deformation; in response to determining that the stability parameters of the joints satisfy the stability compensation condition, the target joint set is updated by performing trajectory optimization deformation; in response to determining that the collision parameters of the joints satisfy the collision avoidance condition, the target joint set is updated by performing spatial offset deformation; the updated target joint set is updated by performing topology consistency verification operation; and the updated target joint set is determined as the compensated joint set.

[0159] Motion smoothing deformation can be achieved using a low-pass filter, such as a Butterworth filter with a cutoff frequency of 5Hz to eliminate sudden velocity changes. Amplitude compensation is triggered when the displacement difference between joints in adjacent frames exceeds a threshold (e.g., 15mm). Trajectory optimization deformation can use a Bezier curve fitting algorithm to smooth segments with trajectory curvature exceeding 0.3, eliminating trajectory jitter. Spatial offset deformation uses a collision detection algorithm to calculate the minimum offset distance; for example, when the joint spacing is less than a preset safety distance (e.g., 3mm), the offset along the normal vector direction is at least 5mm. Topology consistency verification uses a graph neural network to verify the skeletal connections, ensuring that joint angle changes do not exceed ±10% of the original topology.

[0160] Specifically, when the detected rate of change of joint velocity exceeds 200 mm per second, the motion smoothing module is activated. It generates transition frames using an interpolation algorithm, for example, inserting three intermediate frames between two frames to reduce the velocity gradient to below 50 mm per second. The trajectory optimization module resamples jittery trajectories of more than five consecutive frames and generates new motion paths using cubic spline interpolation, controlling the trajectory curvature variance to within 0.05. The spatial offset module detects collision areas using a hierarchical bounding box algorithm and applies repulsive forces to joints that have penetrated the target joint, for example, setting rigid body collision parameters to a friction coefficient of 0.3 and an elastic coefficient of 0.5. The topology verification module compares the deviation of the bone length from the original data frame by frame. When the cumulative deviation exceeds 2%, topology reconstruction is triggered, and the joint positions are corrected using an inverse kinematics algorithm. These processing steps are executed in the order of priority: amplitude compensation, trajectory optimization, and collision avoidance. The final output of the compensated joint set maintains the original motion characteristics while reducing the deformation error of the clothing model to within 3 mm and improving motion continuity by more than 40%.

[0161] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0162] First, motion amplitude parameter analysis is performed on the target keypoint set. Specifically, motion amplitude parameters can be obtained by calculating the change in keypoint position between adjacent frames. For example, a threshold can be set; when the motion amplitude of a keypoint exceeds this threshold, it is determined that the amplitude compensation condition is met.

[0163] Secondly, smooth motion deformation is performed on the joints that meet the amplitude compensation conditions. This step can use interpolation algorithms, such as linear interpolation or Bézier curve interpolation, to add transition frames between adjacent frames, making the motion trajectory smoother.

[0164] Next, a stability analysis is performed on the target key set. Stability can be evaluated by calculating the variance of key position positions within a certain time window. If the variance exceeds a preset threshold, the stability compensation condition is considered met.

[0165] For joints that meet the stability compensation conditions, trajectory optimization deformation is performed. This can be achieved using algorithms such as Kalman filtering, which optimizes the motion trajectory of joints through prediction and correction, thereby improving stability.

[0166] Next, collision detection is performed. Simplified models such as bounding boxes or spheres can be used to quickly detect whether there is a collision between joints. If a collision is detected, it is determined that the collision avoidance conditions are met.

[0167] For joints that meet the collision avoidance conditions, spatial offset deformation is performed. This can be achieved by applying a small displacement in the collision direction to avoid penetration between joints.

[0168] Finally, a topological consistency check is performed on the updated target joint set. This step ensures that the connections and relative positions between joints remain unchanged by comparing the skeletal structure before and after the update. If a change in the topology is detected, the original topological relationships can be restored by adjusting the joint positions.

[0169] After the above steps, the resulting compensated joint set maintains the smoothness and stability of motion, avoids collision problems, and ensures the consistency of the skeletal topology.

[0170] This application further proposes that the fitting request is a static fitting request, and the 3D topology data is static body shape scanning data; key point integrity verification processing is performed on the 3D topology data to obtain verification results, including body shape topology verification processing of the static body shape scanning data, and the topology verification results are used as verification results; based on the 3D topology data, the parameters of the adapted clothing model are constructed, including noise point filtering operation on the static body shape scanning data to obtain a filtered point cloud; curvature distribution calculation is performed on the filtered point cloud to obtain a curvature distribution map; the curvature distribution map is partitioned to obtain a body shape partition sequence; for each partition in the body shape partition sequence, the following steps are performed: determine the partition The system calculates the curvature variance value corresponding to each region; determines the surface area ratio corresponding to each region; in response to the determination that the curvature variance value and surface area ratio meet the preset anomaly conditions, removes regions from the body shape region sequence to update the body shape region sequence; performs feature extraction operations on each region in the updated body shape region sequence to obtain body shape feature vectors; for any two obtained body shape feature vectors, determines the similarity parameter between them; selects the similarity parameter that meets the preset similarity conditions from the determined similarity parameters as the target similarity parameter; and constructs the two body shape feature vectors corresponding to the target similarity parameter and the body shape classification label as the parameters for the clothing adaptation model.

[0171] The noise point filtering operation can employ a statistical outlier filtering algorithm. For example, by calculating the standard deviation of the neighborhood distance of each point in the point cloud, points exceeding three times the standard deviation are identified as noise points. Curvature distribution calculation can be based on a weighted combination of Gaussian curvature and mean curvature. For instance, setting the weight of Gaussian curvature to 0.6 and the weight of mean curvature to 0.4 creates a composite curvature index. Body shape partitioning can utilize a region growing algorithm based on curvature similarity, where seed points are selected from the top 20% of points in the overall distribution based on their curvature values. The calculation of curvature variance can be limited to the set of curvature values ​​of all points within a partition, for example, using an unbiased estimation formula to calculate the sample variance. The determination of surface area proportion can be achieved by calculating the ratio of the surface area of ​​a partition in 3D space to the overall body surface area using a projection method. For example, a region is considered low-influence when this ratio is below 5%. Preset anomaly conditions can be set as a logical AND combination of a curvature variance value exceeding a threshold of 0.25 and a surface area proportion below a threshold of 3%. Feature extraction operations can include multi-dimensional feature fusion, including curvature distribution histograms, centroid coordinates of spatial locations, and the cosine of the angle between normal vectors. The similarity parameter can be calculated using a linear combination of cosine similarity and Euclidean distance, for example, with a cosine similarity weight of 0.7 and an Euclidean distance weight of 0.3. Preset similarity criteria can be set as a threshold filtering condition where the similarity parameter is greater than 0.85.

[0172] Specifically, in a static fitting scenario, outliers in the scanned data are first eliminated using a statistical filtering algorithm, such as using a spherical neighborhood with a radius of 0.5cm for outlier detection. The filtered point cloud is then input into a curvature calculation module, where a curvature distribution map is generated using a differential geometry algorithm. The curvature value of each vertex is calculated using the rate of change of the normal vector of adjacent triangular facets. When dividing the body surface into regions based on the curvature distribution map, an improved K-means clustering algorithm is used to divide the body surface into 8-12 partitions with similar curvature characteristics. The curvature variance of each partition is obtained by calculating the dispersion of the curvature values ​​of all vertices within that partition; for example, partitions with a variance higher than 0.3 are identified as abnormal fluctuation regions. The surface area ratio is determined by calculating the proportion of the partition's surface area to the overall body surface area; for example, partitions with a ratio lower than 2% are marked as redundant regions. When both the curvature variance exceeds the limit and the surface area ratio is too low, the partition is dynamically removed. The updated partition sequence enters the feature extraction stage. Each partition's feature vector consists of three parts: mean curvature, surface area ratio, and centroid coordinates, for example, using a 32-dimensional vector for feature encoding. The similarity calculation between feature vectors employs an improved similarity metric function, such as normalizing and weighting cosine similarity and Manhattan distance. The selected high-similarity feature vector combinations, along with body type classification labels, constitute the adaptation parameters; for example, feature pairs with similarity higher than 0.9 are used as core parameters. Through this technical solution, dynamic removal of abnormal partitions reduces feature extraction error by approximately 37%, and feature similarity screening reduces parameter redundancy by 42%, ultimately improving the size adaptation accuracy of virtual try-on to over 93%.

[0173] As a preferred embodiment, the specific implementation of this application is as follows: In a static fitting request scenario, static body shape scanning data is collected by a 3D laser scanner. The original point cloud data is filtered for noise points using a voxel grid filtering algorithm. After filtering, the principal curvature of the point cloud is calculated using a local surface fitting method to generate a curvature distribution map. The curvature distribution map is partitioned using a region growing algorithm. The curvature variance of each partition in the initial partition sequence is obtained by calculating the standard deviation of the principal curvature of all points within that partition. The surface area ratio is determined by the ratio of the partition area to the overall body shape surface area. A preset abnormal condition is set to trigger a removal operation when the curvature variance exceeds a threshold of 0.15 and the surface area ratio is less than 5%. The updated body shape partition sequence extracts curvature features, area features, and spatial coordinate features using principal component analysis to form a 128-dimensional body shape feature vector. The similarity parameter of any two feature vectors is calculated using a cosine similarity algorithm. A preset similarity condition is set to when the similarity parameter is greater than 0.85. The corresponding feature vector and body shape classification label are then input into a multilayer perceptron model to generate parameters for the adapted clothing model.

[0174] This application further proposes that after controlling the rendering of the virtual fitting effect on the virtual fitting platform, the method also includes: extracting the set of material physical properties corresponding to the target clothing model from the clothing knowledge graph; constructing a fabric dynamics simulation constraint matrix based on the set of material physical properties; and performing physical simulation rendering processing on the virtual fitting effect based on the fabric dynamics simulation constraint matrix to generate a physical simulation fitting effect.

[0175] The material's physical properties set can include parameters such as elastic modulus, coefficient of friction, and bending stiffness. These parameters are stored and accessed through a structured data layer of the clothing knowledge graph. The construction of the fabric dynamics simulation constraint matrix can be achieved by converting the elastic modulus into a deformation constraint equation and the coefficient of friction into a contact force calculation coefficient. For example, the elastic modulus can be set to a range of 0.1-10 GPa, and the coefficient of friction to a range of 0.2-0.8. Physical simulation rendering can employ a mass-spring model or the finite element method, calculating the dynamic interaction between the fabric and the human body model in real time. The distance between mass points can be set to 0.5-2 mm to simulate wrinkle details. Data coupling can be achieved between the material's physical properties set and the motion feature vectors in the adapted clothing model parameters. For example, the joint motion amplitude parameters can be correlated with the fabric elastic parameters to maintain a proportional relationship between the dynamic deformation amplitude and the human body's motion amplitude.

[0176] Specifically, after generating the virtual try-on effect, the system first extracts the set of material physical properties corresponding to the target clothing model from the clothing knowledge graph. The elastic modulus is used to control the stiffness coefficient of the fabric's tensile deformation, and the friction coefficient is used to calculate the sliding resistance when the fabric contacts human skin. Then, a fabric dynamics simulation constraint matrix is ​​constructed based on the material properties. The elastic modulus is transformed into a stiffness coefficient matrix in the deformation constraint equation, and the friction coefficient is transformed into a friction coefficient matrix in the contact force calculation. Based on this constraint matrix, a physics engine performs dynamic calculations on the virtual try-on effect. For example, when the human model performs a turning motion, the stiffness coefficient matrix is ​​used to calculate the drooping deformation of the fabric due to centrifugal force, while the friction coefficient matrix is ​​used to simulate the sliding friction effect between the fabric and the shoulder skin. In the final physically simulated try-on effect, the clothing model not only presents a visual fit but also dynamically displays the natural wrinkle distribution of the fabric due to gravity and the elastic deformation process caused by limb movement. By combining the structured physical parameters of the knowledge graph with the dynamic constraint matrix, the virtual try-on effect can accurately reflect the real physical behavior of the clothing material in dynamic scenes.

[0177] As a preferred embodiment, the solution of this application is implemented as follows: After basic rendering is completed on the virtual fitting platform, the system accesses the clothing knowledge graph database to extract the target clothing material attribute set, which includes elastic modulus, Poisson's ratio, areal density, and shear stiffness parameters. Subsequently, the material attributes are transformed into dynamic constraint equations using the finite element analysis method, constructing a fabric dynamics simulation constraint matrix that includes a mass matrix, stiffness matrix, and damping matrix. The physics engine performs real-time calculations based on the constraint matrix, calculating the displacement and velocity changes of the fabric vertices using an implicit integration algorithm, while simultaneously handling the contact force between the clothing and the human body model using a collision detection module, ultimately generating a physically simulated fitting effect that includes dynamic wrinkles and draping effects.

[0178] This application further proposes to call a security verification engine to perform digital signature verification calculation on the target clothing model; in response to successful verification, to calculate the vertex deformation weight matrix of the target clothing model based on the motion feature vector and body shape feature vector in the parameters of the adapted clothing model; and to generate the migration result based on the vertex deformation weight matrix.

[0179] The security verification engine can integrate hash algorithms or asymmetric encryption algorithms for digital signature verification, such as using the SHA-256 algorithm to verify the integrity of model files. The motion feature vector can include joint motion amplitude parameters and stability parameters, where the motion amplitude parameter is calculated as the ratio of the sum of the magnitudes of the displacement differences between adjacent joints to the total number of joints. The body shape feature vector can be generated based on the partition features in the curvature distribution map, for example, partitions with a curvature variance value below 0.05 and a surface area ratio exceeding 15% are considered valid feature regions. During the calculation of the vertex deformation weight matrix, the spatiotemporal characteristic parameters in the motion feature vector and the spatial morphological parameters in the body shape feature vector are processed through a weighted fusion method, where the weight coefficients for motion parameters can be set in the range of 0.6-0.8, and the weight coefficients for body shape parameters are in the range of 0.2-0.4. During the generation of the transfer results, the weight matrix is ​​applied to the linear transformation operation of the vertex coordinates of the clothing model, for example, using matrix multiplication to achieve precise control of the vertex position offset.

[0180] Specifically, when performing digital signature verification through the security verification engine, the hash value of the target clothing model's metadata is first extracted and compared with the pre-stored security certificate. Once verification is successful, the joint motion trajectory data recorded in the motion feature vector and the curvature distribution features in the body shape feature vector are input to the weight calculation module. During weight calculation, the motion amplitude parameter in the motion feature vector is used to determine the tensile strength of the clothing model during dynamic deformation, while the stability parameter controls the smooth transition during deformation. The curvature distribution features in the body shape feature vector are converted into constraint boundary conditions for vertex deformation; for example, regions with curvature exceeding 0.3 are set as deformation suppression zones. The vertex deformation weight matrix is ​​generated by orthogonally decomposing and recombining the motion parameters and body shape parameters, where dynamic parameters are integrated along the time axis and static parameters are interpolated in the spatial coordinate system. The final weight matrix is ​​applied to update the position of the clothing model's mesh vertices, ensuring that clothing deformation matches both the spatiotemporal changes of human movement and the static features of the body shape. This dual verification and fusion calculation mechanism effectively solves the problem of synergistic optimization between model security and deformation accuracy, while avoiding deformation distortion caused by single parameter calculation in traditional methods.

[0181] As a preferred embodiment, the solution of this application is implemented as follows: When performing clothing model migration and adaptation deformation processing, the target clothing model is first verified by a digital signature through a security verification engine. Specifically, a digital signature verification mechanism based on the RSA algorithm is adopted, wherein the security verification engine calls the public key certificate pre-installed in the virtual fitting platform to decrypt the hash digest carried by the target clothing model and compares it with the hash value calculated in real time. If the hash values ​​match, the verification is deemed successful. After successful verification, motion feature vectors and body shape feature vectors are extracted from the parameters of the adapted clothing model. The motion feature vectors consist of acceleration and displacement parameters of the human joint motion trajectory, and the body shape feature vectors consist of curvature gradient parameters in the body shape scanning data. Subsequently, the two feature vectors are input into a multilayer perceptron for weighted fusion calculation, and a vertex deformation weight matrix is ​​output. Each element in this matrix corresponds to the deformation influence coefficient of the clothing model mesh vertex. The weight matrix is ​​further applied to the vertex coordinates of the target clothing model through a linear hybrid skinning algorithm to generate the migrated clothing model data as the migration result.

[0182] Through the above technical solutions, this application effectively solves the problems of data security risks and insufficient deformation accuracy during the migration process of clothing models. A digital signature verification mechanism blocks unauthorized tampering of the model during migration, preventing the abuse of system resources caused by malicious code injection. The joint calculation mechanism of motion feature vectors and body shape feature vectors enables the vertex deformation weight matrix to accurately reflect the dual influence characteristics of human dynamic movement and static body shape. In strong deformation scenarios with knee bending angles exceeding 120 degrees, the vertex displacement error of the clothing model mesh is controlled within 2 millimeters, improving the matching accuracy between clothing fold shapes and human movements. The weight matrix generation method based on physical constraints avoids the fabric penetration phenomenon caused by traditional linear interpolation algorithms, improving the realism of dynamic fitting effects by more than 35%.

[0183] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for automated transfer of clothing models and virtual try-on, comprising: In response to receiving a fitting request for a target garment through a virtual fitting platform, feature extraction is performed on the fitting request to obtain the target human body model parameters corresponding to the fitting request; Collect the three-dimensional topological data of the target human body model; Based on the three-dimensional topology data, construct the parameters of the adapted clothing model; A virtual fitting interface is displayed that corresponds to the parameters of the adapted clothing model, wherein the virtual fitting interface integrates a parameter adjustment module; The parameters of the adapted clothing model are updated according to the material parameters of the corresponding parameter adjustment module. In response to obtaining a target clothing model through the virtual fitting platform, the target clothing model is subjected to migration adaptation deformation processing based on the updated adaptation clothing model parameters to obtain a migration result; In response to determining that the migration result represents a successful adaptation, the virtual fitting platform is controlled to render the virtual fitting effect.

2. The method according to claim 1, wherein, The target human body model parameters include body type classification identifiers and model topology identifiers; Before acquiring the three-dimensional topological data of the target human body model, the method further includes: In response to the detection of a confirmation command applied to the fitting request interface, skeletal joint data corresponding to the target human body model are collected according to the model topology identifier, wherein the fitting request interface corresponds to the fitting request and the body type classification identifier is displayed in the fitting request interface; Determine whether the skeletal joint data meets a preset complexity condition; in response to determining that the skeletal joint data meets the preset complexity condition, assign the corresponding... The feature extraction command for the fitting request is sent to the model analysis engine; and The acquisition of the three-dimensional topological data of the target human body model includes: In response to receiving the feature extraction completion response returned by the model analysis engine, the three-dimensional topological data of the target human body model is collected.

3. The method according to claim 2, wherein, The step of constructing the parameters for the adapted clothing model based on the three-dimensional topology data includes: The three-dimensional topology data is subjected to key point integrity verification processing to obtain verification results; in response to determining that the verification results represent the integrity of key points, the parameters of the adapted clothing model are constructed based on the three-dimensional topology data.

4. The method according to claim 3, wherein, The fitting request is a dynamic fitting request, and the three-dimensional topology data is temporal motion capture data; as well as The key point integrity verification process for the three-dimensional topology data, to obtain the verification result, includes: The motion continuity analysis of the time-series motion capture data is performed to obtain the motion continuity results as the verification results.

5. The method according to claim 4, wherein, The step of constructing the parameters for the adapted clothing model based on the three-dimensional topology data includes: The temporal motion capture data is subjected to motion decomposition operations to obtain a sequence of motion frames; for each motion frame in the sequence of motion frames, the following steps are performed: Determine whether a complete skeletal joint exists in the action frame; In response to determining that there are complete skeletal joints in the action frame, the coordinates of the joints in the action frame are normalized to obtain a standard joint set. Motion amplitude analysis is performed on the standard joint set to obtain joint motion amplitude parameters; Stability analysis was performed on the standard joint set to obtain joint stability parameters; Collision detection calculations are performed on the standard set of joints to obtain joint collision parameters; Based on the joint motion amplitude parameters, the joint stability parameters, and the joint collision parameters, construct the motion frame parameters; Based on the obtained parameters of each action frame, select the target set of joints that meet the preset action conditions from the obtained set of standard joints. Based on the joint motion amplitude parameters, joint stability parameters and joint collision parameters corresponding to the target joint set, motion compensation calculation is performed on the target joint set to obtain the compensated joint set. The compensated joint set is subjected to feature encoding to obtain motion feature vectors; based on the motion feature vectors and the body type classification identifier, the parameters of the adapted clothing model are constructed.

6. The method according to claim 5, wherein, The motion amplitude analysis of the standard joint set to obtain joint motion amplitude parameters includes: The velocity vector of the standard joint set is calculated to obtain the joint velocity matrix; the joints in the standard joint set that satisfy the preset static threshold are removed to obtain the dynamic joint set. For each joint in the dynamic joint set, the following steps are performed: the joints adjacent to the joint in the joint velocity matrix are determined as adjacent joints; A first displacement difference and a second displacement difference are generated based on the displacement vectors of the adjacent joints and the displacement vectors of the joints. The sum of the magnitudes of the first displacement differences and the second displacement differences is determined as the third displacement difference. The number of each joint in the dynamic joint set is determined as the total number of joints. The ratio of the third displacement difference to the total number of joints is determined as the joint motion amplitude parameter corresponding to the action frame.

7. The method according to claim 5, wherein, The step of performing motion compensation calculations on the target joint set based on the joint motion amplitude parameters, joint stability parameters, and joint collision parameters corresponding to the target joint set to obtain a compensated joint set includes: In response to determining that the joint motion amplitude parameters corresponding to the target joint set meet the amplitude compensation condition, the target joint set is subjected to motion smoothing deformation to update the target joint set; In response to determining that the joint stability parameters corresponding to the target joint set meet the stability compensation condition, the target joint set is subjected to trajectory optimization deformation to update the target joint set; In response to determining that the collision parameters of the target joint set satisfy the collision avoidance conditions, the target joint set is spatially offset and deformed to update the target joint set. Perform a topology consistency check operation on the updated target key set to update the target key set; The updated target set of joints is determined as the compensated set of joints.

8. The method according to claim 3, wherein, The fitting request is a static fitting request, and the three-dimensional topology data is static body shape scan data; The process of performing key point integrity verification on the three-dimensional topology data to obtain verification results includes: Perform body topology verification processing on the static body shape scan data, and obtain the topology verification result as the verification result; and The step of constructing the parameters for the adapted clothing model based on the three-dimensional topology data includes: The static body shape scan data is subjected to noise point filtering operation to obtain the filtered point cloud; The curvature distribution of the filtered point cloud is calculated to obtain a curvature distribution map; The curvature distribution map is partitioned to obtain a body shape partition sequence; For each partition in the body size partition sequence, perform the following steps: Determine the curvature variance value corresponding to the partition; Determine the surface area percentage corresponding to the partition; In response to determining that the curvature variance value and the surface area ratio meet preset abnormal conditions, the partition is removed from the body shape partition sequence to update the body shape partition sequence; Feature extraction is performed on each partition in the updated body shape partition sequence to obtain a body shape feature vector; For any two obtained body shape feature vectors, determine the similarity parameter between the two body shape feature vectors; Select the similarity parameter that meets the preset similarity conditions from the determined similarity parameters as the target similarity parameter; The two body shape feature vectors corresponding to the target similarity parameter and the body shape classification identifier are used to construct the parameters of the clothing adaptation model.

9. The method according to claim 1, wherein, After controlling the virtual fitting platform to render the virtual fitting effect, the method further includes: Extract the set of material physical properties corresponding to the target clothing model from the clothing knowledge graph; Construct a constraint matrix for fabric dynamics simulation based on the set of physical properties of the material; Based on the fabric dynamics simulation constraint matrix, the virtual try-on effect is physically simulated and rendered to generate a physically simulated try-on effect.

10. The method according to claim 9, wherein, The step of performing migration adaptation deformation processing on the target clothing model based on the updated adapted clothing model parameters to obtain the migration result includes: The security verification engine is invoked to perform digital signature verification calculations on the target clothing model. In response to successful verification, based on the motion feature vector in the parameters of the adapted clothing model... Using the body shape feature vector, calculate the vertex deformation weight matrix of the target clothing model; The migration result is generated based on the vertex deformation weight matrix.

Images