Virtual fitting fitness evaluation method based on dynamic posture

Abstract

The invention relates to the technical field of virtual fitting, in particular to a virtual fitting fit evaluation method based on dynamic postures, which comprises the following steps: acquiring dynamic posture data when a user executes a preset action sequence, and constructing a digital fabric physical model containing physical parameters of tensile rigidity, bending rigidity and shearing rigidity, the stretching, wrinkling and fitting changes of the clothes are calculated under the driving of the dynamic posture data, and dynamic deformation data are generated; then, initial gap distribution and initial pressure mapping between the clothes and the body are calculated based on the dynamic deformation data, and stable gap distribution and accurate pressure mapping are obtained through space-time consistency analysis; and finally, generating a fitness evaluation index including a dynamic fitness index, a local compression coefficient and an overall fitness score, and outputting a fitness report through visual rendering. According to the method, the fitting state of the clothes under dynamic movement can be truly reflected, and the fine evaluation capability of virtual fitting is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of virtual fitting, and in particular to a virtual fitting fit evaluation method based on dynamic posture. BACKGROUND

[0002] With the development of virtual reality, three-dimensional reconstruction and computer graphics technology, virtual fitting technology has gradually become an important tool for clothing e-commerce, clothing customization and online code selection. Through three-dimensional human modeling and cloth simulation, users can view the appearance and general fit of clothes without actually dressing. However, traditional virtual fitting relies on static posture and single-frame human models, which cannot reflect the dynamic fitting state of clothes caused by changes in user actions in actual dressing scenarios, such as shoulder displacement when walking, armpit compression when raising arms, or side waist tension when twisting, etc. At the same time, the stretching, bending and shearing of clothing materials will significantly affect the wrinkles, tightness and compression of clothes during dynamic motion, so more detailed dynamic simulation technology is needed to realistically present the dressing effect.

[0003] Most existing virtual fitting technologies have problems such as inability to accurately capture user dynamic posture, inability to simulate real cloth deformation under complex motion, and inability to provide quantitative fit indicators. On the one hand, existing technologies generally use simplified skeletal models or posture estimation based solely on depth maps, which cannot obtain high-precision body joint positions and motion trajectories, resulting in inaccurate clothing deformation simulation. On the other hand, existing cloth simulation algorithms rely on static collision detection and rough physical approximation, which cannot maintain energy conservation under rapid motion and cannot avoid clothes penetrating the human model. In addition, existing solutions generally lack quantitative analysis of the gap distribution and pressure mapping between clothes and the human body, cannot provide dynamic fitting indices, local compression coefficients and other evaluation indicators, and cannot automatically generate structured fit reports and visual presentations based on evaluation results, making it difficult for users to intuitively understand dressing problems and limiting the ability of clothing design optimization and personalized adjustment. SUMMARY

[0004] The present application provides a virtual fitting fit evaluation method based on dynamic posture, which realizes accurate quantitative evaluation of the dynamic fit performance of clothes by acquiring dynamic posture data, simulating dynamic deformation of clothes, calculating gap and pressure distribution and generating visual fit reports.

[0005] A virtual fitting fit evaluation method based on dynamic posture, comprising the following steps:

[0006] S1: acquiring dynamic posture data of a user, the dynamic posture data including body joint positions and motion trajectories of the user when performing a preset action sequence;

[0007] S2: simulate the dynamic behavior of the clothes on the user based on the dynamic pose data, generate dynamic deformation data of the clothes, wherein the dynamic behavior includes stretching, wrinkling and fitting changes of the clothes;

[0008] S3: evaluate the fit of the clothes according to the dynamic deformation data of the clothes, generate fit evaluation indicators, and the fit evaluation indicators include gap distribution and pressure mapping between the clothes and the body;

[0009] S4: output the fit evaluation indicators to form a visual fit report, and the report includes fit scores under dynamic poses and improvement suggestions.

[0010] Optionally, the S1 includes:

[0011] S11: obtain a preset action sequence, the preset action sequence includes a plurality of standard action instructions for guiding the user to perform a dynamic pose, wherein the preset action sequence is provided to the user through a user interface display or voice prompt, and a signal is received to confirm the start of the user, to generate an executable preset action sequence;

[0012] S12: based on the preset action sequence, real-time capture multiple frames of motion data of the user performing the preset action sequence through a depth camera or an inertial measurement unit sensor, wherein the multiple frames of motion data include three-dimensional point cloud information or acceleration and angular velocity data of the user's body in time sequence, and output the multiple frames of motion data;

[0013] S13: calculate the body joint position and motion trajectory of the user from the multiple frames of motion data, wherein the body joint position is extracted from the multiple frames of motion data through a skeleton tracking algorithm or a machine learning model, and the motion trajectory is derived by analyzing the change of the body joint position in the time dimension, and the body joint position and the motion trajectory are output;

[0014] S14: generate the dynamic pose data according to the body joint position and the motion trajectory, wherein the dynamic pose data includes a body joint position matrix and a motion trajectory vector, which are used as input for subsequent steps.

[0015] Optionally, the executable preset action sequence includes: selecting and combining a personalized preset action sequence from a predefined action library according to the user's body type basic data or historical fitting records, wherein the body type basic data includes height, weight and body type classification.

[0016] Optionally, the dynamic pose data further includes: data standardization processing of the body joint position and the motion trajectory, eliminating abnormal frame data in the execution process of the preset action sequence, and completing the missing body joint position to form complete and consistent dynamic pose data.

[0017] Optionally, S2 comprises:

[0018] S21: constructing a digital fabric physical model of the garment, the digital fabric physical model containing physical parameters of tensile stiffness, bending stiffness and shear stiffness of the fabric, and generating the digital fabric physical model based on the physical parameters;

[0019] S22: coupling the digital fabric physical model with the dynamic pose data obtained from step S1, calculating real-time deformation of the digital fabric physical model under the action of the dynamic pose data through a physics engine, and generating preliminary deformation data of the garment, wherein the real-time deformation includes stretch, wrinkle and fit changes of the garment;

[0020] S23: correcting the preliminary deformation data of the garment for dynamic physical effects, correcting the non-penetration constraint relationship between the garment and the body based on the energy conservation principle and cloth collision detection algorithm, and outputting the corrected dynamic deformation data of the garment, the dynamic deformation data including accurate stretch distribution, wrinkle depth and fit change amount.

[0021] Optionally, the construction of the digital fabric physical model of the garment comprises: obtaining physical sample data of the target fabric through a fabric mechanical tester, and inputting the physical sample data into a material parameter inversion algorithm to calibrate the physical parameters of tensile stiffness, bending stiffness and shear stiffness in the digital fabric physical model.

[0022] Optionally, S3 comprises:

[0023] S31: based on the dynamic deformation data of the garment obtained from step S2, calculating the signed distance field between the garment mesh model and the user body mesh model to generate an initial gap distribution, and calculating the initial pressure mapping through a predefined pressure-stress conversion function according to the stress tensor in the dynamic deformation data of the garment;

[0024] S32: performing spatiotemporal consistency analysis on the initial gap distribution and the initial pressure mapping, smoothing inter-frame mutations through a time series filter, and completing data missing caused by occlusion through a spatial interpolation algorithm to form a stable gap distribution and an accurate pressure mapping;

[0025] S33: according to the stable gap distribution and the accurate pressure mapping, comprehensively calculating a dynamic fit index, a local compression coefficient and an overall fit score, and integrating to generate a final fit evaluation index, wherein the fit evaluation index includes quantized gap distribution values and pressure mapping values.

[0026] Optionally, the signed distance field between the garment mesh model and the user body mesh model comprises: establishing a spatial distance query data structure based on the user body mesh model, traversing the shortest directed distance from each vertex of the garment mesh model to the user body mesh model, wherein a positive value represents a gap and a negative value represents a penetration, thereby generating the initial gap distribution.

[0027] Optionally, the S4 comprises:

[0028] S41: parse the obtained fit evaluation indicators, structure the dynamic fit index, local compression coefficient, overall fit score, gap distribution value and pressure mapping value in the fit evaluation indicators according to a predefined report template, and generate structured fit data;

[0029] S42: based on the structured fit data, automatically generate a preliminary fit report containing a comprehensive fit score and specific improvement suggestions, wherein the comprehensive fit score is calculated by weighting and fusing each value in the structured fit data, and the specific improvement suggestions are generated by matching abnormal gap distribution values and abnormal pressure mapping values in the structured fit data with a pre-stored garment modification knowledge base through a rule engine;

[0030] S43: visually render the preliminary fit report, mark the comprehensive fit score in the form of stars, and superimpose and fuse the specific improvement suggestions with the key frame image of the user's dynamic posture, and output the final visual fit report through an augmented reality interface.

[0031] Optionally, the structured processing of the fit evaluation indicators according to the predefined report template comprises: data integrity verification and outlier filtering on the gap distribution value and the pressure mapping value in the fit evaluation indicators, and storing the verified data in the corresponding data structure according to the body region partition, to generate complete and standardized structured fit data.

[0032] The beneficial effects of the present application are:

[0033] The present application can accurately represent the three-dimensional posture change of the user under different action states by obtaining dynamic posture data containing body joint position matrix and motion trajectory vector. Compared with the traditional fitting method which only relies on static posture or simplifies the skeletal model, the present application can dynamically capture the real posture of the user in complex actions such as walking, arm lifting and rotating. Combined with the digital fabric physical model containing physical parameters of tensile stiffness, bending stiffness and shear stiffness, the stretching, wrinkling and fitting changes of the clothes can be simulated under dynamic driving conditions, and the clothing deformation is corrected through the energy conservation principle and cloth collision detection algorithm, so that the dynamic deformation data is more consistent with the physical behavior of the real fabric, thereby significantly improving the accuracy and reliability of the fit evaluation.

[0034] The present application can obtain an initial gap distribution that can distinguish between gaps and penetrations by calculating a signed distance field through a spatial distance query data structure, and generate an initial pressure map using a pressure-stress conversion function, thereby realizing real quantification of the gap and compression relationship between the clothes and the user's body. Then, the Kalman filter is used to smooth the time series mutations, and the radial basis function interpolation method is used to complete the data missing caused by occlusion, so that the gap distribution and the pressure map have continuity, stability and integrity in time and space. Through the above multi-level spatio-temporal consistency processing, the present application can provide high-precision and reliable gap and pressure data at key positions (such as shoulders, bust, waist, thighs, etc.), providing a solid data foundation for dynamic fitting index, local compression coefficient and overall fit score.

[0035] The present application structures the gap distribution value, the pressure mapping value, the dynamic fitting index, the local compression coefficient and the overall fit score according to the pre-defined template, and calculates the comprehensive fit score by weighted fusion, and automatically retrieves the version adjustment scheme from the clothing modification knowledge base to generate specific improvement suggestions through the rule engine, so that the final output fit report not only contains quantitative indicators, but also gives clear optimization direction. In addition, through WebGL technology and shader program, the gap distribution and pressure map are superimposed and rendered into the user's body model in the form of a heat map, and are superimposed and fused with the dynamic posture key frame image, and finally output through the augmented reality interface, so that the user can intuitively view the location and change trend of the fit problem. The present application has significant advantages in user experience and clothing adjustment guidance, and can be widely applied to virtual fitting, e-commerce code selection, clothing customization and other scenes. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only a part of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort.

[0037] Fig. 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0038] Fig. 2 This is a schematic diagram of the S1 process in an embodiment of the present invention. Detailed Implementation

[0039] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0040] It should be noted that the use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.

[0041] Generally, terms can be understood at least partly from their use in context. For example, depending at least partly on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or a combination of features, structures, or characteristics in a plural sense. Additionally, the term "based on" can be understood not necessarily to convey an exclusive set of factors, but rather, alternatively, depending at least partly on the context, to allow for the presence of other factors that are not necessarily explicitly described.

[0042] like Figs. 1-2 As shown, a virtual fitting fit assessment method based on dynamic posture includes the following steps:

[0043] S1: Obtain the user's dynamic posture data, which includes the user's body joint positions and movement trajectories when performing a preset action sequence, specifically:

[0044] S11: Displays multiple standard action commands to the user through a pre-defined human-computer interaction interface. These commands are displayed graphically one by one on the interface, while a voice prompt module outputs a description of each command. First, the system reads the user's basic body data, including height, weight, and body type, and then extracts previously performed actions from stored historical fitting records.

[0045] According to the body type basic data and the historical fitting record, a plurality of standard action instructions are selected from a predefined action library according to a preset action matching rule, and are combined in a predetermined order to form a personalized preset action sequence.

[0046] After the personalized preset action sequence is formed, all action contents of the preset action sequence are completely listed on the interface, and a voice prompt is played synchronously to prompt the action sequence and action requirements to be performed by the user.

[0047] When the user confirms to start performing through a confirmation button on the interface or a voice confirmation instruction, the personalized preset action sequence is determined as an executable preset action sequence.

[0048] Subsequently, the executable preset action sequence is recorded in time sequence for guiding the collection of subsequent multi-frame motion data.

[0049] S12: According to the executable preset action sequence, a depth camera and at least one inertial measurement unit sensor are started to collect multi-frame motion data. The depth camera continuously collects three-dimensional point cloud information at each time frame by continuously collecting the whole process of the user performing the preset action sequence at a preset frame rate. The inertial measurement unit sensor is worn on a designated part of the user's body to collect acceleration data and angular velocity data at the same time interval as the depth camera.

[0050] In the collection process, to ensure the consistency of the multi-frame motion data, the three-dimensional point cloud information collected by the depth camera and the acceleration data and angular velocity data collected by the inertial measurement unit sensor are strictly aligned according to the time stamp. The three-dimensional point cloud information, the acceleration data and the angular velocity data after alignment are calibrated and integrated by using a preset sensor fusion algorithm, the coordinate systems of different sensors are unified to the same reference coordinate system through a coordinate transformation step, and weighted fusion is performed according to the noise characteristics of the sensors.

[0051] Through the above steps, the fused motion information at each time frame is obtained, and the fusion results of all time frames are arranged in time sequence to form multi-frame motion data.

[0052] The multi-frame motion data contains user motion information in all time periods corresponding to the preset action sequence, which is used to calculate the body joint position and motion trajectory in step S13.

[0053] S13: The multi-frame motion data obtained in S12 is read, and the body joint position of the user is calculated from the multi-frame motion data by using a machine learning model.

[0054] The human body structure is recognized by using a pre-trained machine learning model. The machine learning model is constructed by using a pose estimation algorithm based on a convolutional neural network, and the body joint positions at each time frame are output by extracting and matching features in multiple frames of motion data.

[0055] After obtaining the body joint positions arranged in time sequence, the motion trajectories of the body joints are calculated according to the changes of the body joint positions in the time dimension.

[0056] Specifically, the body joint positions of adjacent time frames are subjected to differential analysis and interpolation processing to obtain the motion paths changing with time, thereby forming continuous motion trajectories. Through the above steps, the body joint positions and motion trajectories corresponding to the execution process of the preset action sequence are obtained.

[0057] After step S13 is completed, the body joint positions and motion trajectories are output as inputs for generating dynamic pose data in step S14.

[0058] S14: Generating dynamic pose data according to the body joint positions and motion trajectories output in step S13.

[0059] First, the body joint positions and motion trajectories are subjected to data standardization processing, which includes normalizing the body joint positions at all time frames according to a unified reference coordinate system, and scaling the amplitudes of the motion trajectories according to a preset amplitude range, so that the body joint positions and motion trajectories of different users can be represented on a unified scale.

[0060] After the data standardization is completed, abnormal frame data generated during the execution of the preset action sequence is detected. The abnormal frame data includes frames in which the body joint positions are severely deviated due to acquisition errors, and frames in which the motion trajectories are discontinuously changed in time. The abnormal frame data is identified by a preset threshold rule, and is excluded from the subsequent processing process.

[0061] After the abnormal frame data is excluded, the missing body joint positions caused by the exclusion of abnormal frames, short-time occlusion or sensor frame loss are completed. In the completion process, the body joint positions at missing time points are calculated by interpolation method according to the body joint positions of adjacent time frames, so as to ensure the continuity of the body joint positions in time during the entire execution of the preset action sequence.

[0062] After the standardization, abnormality exclusion and missing completion processes are completed, the body joint positions at all time frames are arranged in time sequence and joint number, and a body joint position matrix is constructed. The motion trajectories of all body joints at each time frame are spliced in a preset order to construct a motion trajectory vector. The body joint position matrix and the motion trajectory vector jointly constitute the dynamic pose data.

[0063] Finally, the dynamic pose data including the body joint position matrix and the motion trajectory vector are taken as the input of step S2 for the subsequent dynamic behavior simulation and dynamic deformation calculation of the garment.

[0064] S2: simulate the dynamic behavior of the garment on the user based on the dynamic pose data, and generate the dynamic deformation data of the garment, wherein the dynamic behavior includes the stretch, wrinkle and fit change of the garment, specifically:

[0065] S21: first, construct a digital fabric physical model of the garment. In the construction process, to ensure that the digital fabric physical model can accurately reflect the real physical characteristics of the target fabric, physical sample data of the target fabric is obtained through a fabric mechanical tester. The fabric mechanical tester applies tensile load, bending load and shear load to the target fabric according to a preset loading mode, and records the mechanical response of the fabric under the action of tensile, bending and shear, respectively.

[0066] The above physical sample data is input into a material parameter inversion algorithm. The material parameter inversion algorithm calibrates the tensile stiffness, bending stiffness and shear stiffness in the model based on the mechanical response characteristics of the target fabric, minimizes the error between the sample data and the model prediction value, and makes the tensile stiffness, bending stiffness and shear stiffness physical parameters in the digital fabric physical model consistent with the real fabric.

[0067] After completing the material parameter inversion, a digital fabric physical model is constructed based on the calibrated tensile stiffness, bending stiffness and shear stiffness physical parameters, and the digital fabric physical model is used for dynamic coupling calculation in step S22.

[0068] The inversion objective function is represented as:

[0069] ;

[0070] wherein, is the error objective function value of the inversion algorithm, which is used to measure the difference between the simulated force-displacement curve and the real test curve, is the number of data points participating in the inversion calculation, the simulated force value calculated by the digital fabric physical model at the displacement point, the real force value measured by the fabric mechanical tester at the displacement point.

[0071] S22: couple the digital fabric physical model constructed in step S21 with the dynamic pose data obtained in step S1. The dynamic pose data includes body joint positions and motion trajectories. By taking the body joint positions as the driving boundary and the motion trajectories as the motion driving force, the digital fabric physical model produces real-time deformation under the action of the dynamic pose.

[0072] To realize real-time deformation calculation, a physics engine based on position dynamics is used to iteratively solve the digital fabric physical model. In each time frame, the physics engine based on position dynamics determines the driving conditions of the clothes mesh according to the body joint positions and motion trajectories, and iteratively updates each vertex in the digital fabric physical model frame by frame. In the iteration process, the mechanical response of the fabric unit is calculated according to the tensile stiffness, bending stiffness and shear stiffness physical parameters in the digital fabric physical model, and the position changes of each vertex are continuously solved, so that the clothes produce stretching, wrinkling and fitting changes under the driving of dynamic poses.

[0073] The iterative update is represented as:

[0074] ;

[0075] wherein, is the three-dimensional position of a vertex of the clothes mesh at the current time stamp, is the updated position of the vertex at the next time stamp, is the time step of the iterative solution, consistent with the sampling frequency of the dynamic pose data, is the resultant acceleration of the vertex at the current time stamp, including the acceleration generated by the stretching force, bending force, shear force and dynamic inertial force.

[0076] After the iteration solution converges, the preliminary deformation data of the clothes is obtained. The preliminary deformation data is the dynamic response result of the clothes under the dynamic pose, including the stretching, wrinkling and fitting changes of the clothes, and serves as the input data of step S23.

[0077] S23: To improve the physical accuracy of the deformation of the clothes, the preliminary deformation data of the clothes obtained in step S22 is corrected for dynamic physical effects.

[0078] First, according to the principle of conservation of energy, the kinetic energy and elastic potential energy generated by the dynamic pose in the preliminary deformation data of the clothes are calculated. According to the distribution relationship between the kinetic energy and the elastic potential energy, the amplitude of the preliminary deformation of the clothes is adjusted, so that the deformation amplitude conforms to the energy distribution law of the real fabric under rapid motion.

[0079] After energy correction, the penetration phenomenon between the clothes vertices and the body mesh model reconstructed based on the dynamic pose data is detected by a cloth collision detection algorithm. In the penetration detection, the spatial distance between the clothes vertices and the body mesh model is calculated to determine whether the mesh overlap occurs. When the penetration between the clothes vertices and the body mesh model is detected, according to the non-penetration constraint rule of the cloth collision detection algorithm, the clothes vertices that have penetrated are moved to a non-penetration position outside the surface of the body mesh model along the normal direction.

[0080] After collision correction, the dynamic deformation data of the corrected clothing is obtained. The dynamic deformation data includes the corrected precise stretch distribution, wrinkle depth, and fit change, and serves as the input for the subsequent step S3.

[0081] S3: Based on the dynamic deformation data of the clothing, assess the fit of the clothing and generate fit assessment indicators. The fit assessment indicators include the distribution of gaps and pressure mapping between the clothing and the body, specifically:

[0082] S31: First, read the dynamic deformation data output in step S2, which includes the stretching distribution, wrinkle depth and fit change of the clothing under dynamic posture.

[0083] After acquiring the dynamic deformation data, a clothing mesh model is established based on the dynamic deformation data of the clothing, and a user body mesh model is reconstructed based on the dynamic posture data generated in step S1.

[0084] To calculate the signed distance field between the clothing mesh model and the user body mesh model, a spatial distance query data structure is established based on the user body mesh model.

[0085] The spatial distance query data structure adopts a three-dimensional spatial acceleration structure based on the user's body mesh model, which is used to efficiently query the shortest directed distance from any vertex in the clothing mesh model to the surface of the user's body mesh model.

[0086] During the distance calculation process, each vertex of the clothing mesh model is traversed, and the shortest directed distance from that vertex to the surface of the user's body mesh model is calculated. A positive value of this shortest directed distance indicates that there is a gap between the clothing and the body, while a negative value indicates that the clothing vertex penetrates the user's body surface. By calculating the shortest directed distances to all clothing mesh vertices, an initial gap distribution between the clothing and the body is generated.

[0087] After calculating the initial gap distribution, the initial pressure mapping is calculated using a predefined pressure-stress transformation function based on the stress tensor in the dynamic deformation data.

[0088] The predefined pressure-stress conversion function maps the tensile and shear stresses contained in the stress tensor to the corresponding force applied per unit area to the user's skin.

[0089] An initial pressure mapping is generated by performing point-by-point transformation on the stress tensor of each vertex of the clothing mesh model.

[0090] The predefined pressure-stress transformation function is expressed as:

[0091] ;

[0092] in, In the initial pressure mapping, the clothing mesh model of the first... Each vertex corresponds to a force per unit area acting on human skin. For the stress tensor in the dynamic deformation data, in the clothing mesh model... Stress tensor at each vertex In the first The unit normal vector at each vertex, determined by the surface normal direction of the user's body mesh model, is used to characterize the direction of pressure application.

[0093] S32: Perform spatiotemporal consistency analysis on the initial gap distribution and initial pressure mapping obtained in step S31 to improve the stability and integrity of the gap distribution and pressure mapping.

[0094] First, a time-series filter is applied to smooth the initial gap distribution and initial pressure mapping across consecutive time frames. The time-series filter uses a Kalman filter, which optimally weights the estimated and predicted values ​​for each time frame to suppress random noise in the data, smooth abrupt changes between frames, and ensure the continuity and consistency of the gap distribution and pressure mapping in the time dimension.

[0095] After completing the temporal smoothing process, a spatial interpolation algorithm is used to fill in the data gaps caused by occlusion. The spatial interpolation algorithm adopts the radial basis function interpolation method. By using the vertices of the clothing mesh in the unoccluded area of ​​the dynamic deformation data as known points, a radial basis function model with the spatial position of the vertices as parameters is constructed, and the gap data and pressure data of the occluded area are reconstructed by interpolation.

[0096] This interpolation method enables the shaded area to obtain consistent and coherent gap and pressure values ​​with the surrounding area, thereby forming a complete and stable gap distribution and accurate pressure mapping.

[0097] S33: Based on the stable gap distribution and accurate pressure mapping obtained in step S32, calculate the final fit evaluation index.

[0098] The fit assessment indicators include dynamic fit index, local pressure coefficient and overall fit score, and include quantified gap distribution value and pressure mapping value.

[0099] When calculating the dynamic fit index, based on a stable gap distribution and accurate pressure mapping, the calculation is performed on the entire time series covered by the preset key body areas over the dynamic posture data.

[0100] For each key body region, the interval uniformity index and pressure comfort index of that region are calculated over the entire time series, and the two are weighted and averaged with preset weights.

[0101] The final weighted average is normalized, and the normalized result is used as the dynamic fit index.

[0102] The weighted average of key body regions is expressed as follows:

[0103] ;

[0104] in, The weighted average index is calculated based on a stable gap distribution and accurate pressure mapping within a pre-defined key body area. The median value of the dynamic fit index before normalization is also included. This refers to the gap uniformity index calculated over the entire time series covered by dynamic attitude data in this key body region. This refers to the pressure comfort index calculated over the entire time series covered by dynamic posture data for this key body region. This is a weighting coefficient for the gap uniformity index, used to characterize the importance of gap distribution in the calculation of the dynamic fit index. This is a weighting coefficient for the pressure comfort index, used to characterize the importance of pressure comfort in the calculation of the dynamic fit index.

[0105] After obtaining the dynamic fit index, it is combined with the local compression coefficient and overall fit score to form the final fit assessment index. This fit assessment index is used to describe the overall fit performance of the garment under dynamic postures, providing basic data for the subsequent generation of visualization reports.

[0106] S4: Based on the fit assessment metrics, generate a visualized fit report, including fit scores under dynamic postures and improvement suggestions, specifically:

[0107] S41: First, analyze the fit assessment indicators obtained in step S3. These indicators include the dynamic fit index, local pressure coefficient, overall fit score, gap distribution value, and pressure mapping value. During analysis, each indicator is formatted according to a predefined report template, ensuring that all indicators are organized according to a unified data type and field structure.

[0108] Subsequently, based on the predefined report template, data integrity verification was performed on the gap distribution values ​​and pressure mapping values ​​in the fit assessment indicators.

[0109] Data integrity verification includes detecting whether data is missing, whether there are frames inconsistent with the dynamic pose sequence, and whether there are abnormal data points that exceed the numerical range.

[0110] The detected outliers are filtered by outlier filtering, and the gap distribution values and pressure mapping values that do not meet the data stability requirements are removed by threshold rules.

[0111] After completing data verification and filtering, the verified gap distribution values and pressure mapping values are divided according to the predefined body regions, and the values in each region are stored in the corresponding data structure.

[0112] Through this regional storage method, complete and standardized structured fit data is generated, and a unified data basis is provided for further analysis in step S42.

[0113] S42: Based on the structured fit data obtained in step S41, a preliminary draft of the fit report containing the comprehensive fit score and specific improvement suggestions is automatically generated.

[0114] First, the comprehensive fit score is obtained by weighted fusion calculation of each value in the structured fit data. The weighted fusion calculation is based on the preset fusion weight, and the dynamic fit index, local compression coefficient, overall fit score and regional gap distribution value and pressure mapping value are comprehensively processed, so that the comprehensive fit score can objectively reflect the overall fit performance of the clothes in the dynamic posture.

[0115] The weighted fusion of the comprehensive fit score is expressed as:

[0116] ;

[0117] Wherein, is the comprehensive fit score, which is obtained by weighted fusion of the structured fit data, is the number of indicators participating in the fusion calculation, including dynamic fit index, local compression coefficient, overall fit score, gap distribution value, pressure mapping value, etc. The first value in the structured fit data, such as the gap distribution value or pressure mapping value of a certain region, is the weight corresponding to the first indicator, which comes from the preset weight configuration.

[0118] Subsequently, the rule engine is called to perform matching analysis on the abnormal gap distribution values and abnormal pressure mapping values in the structured fit data.

[0119] The rule engine matches the numerical value of each body region in the structured fit data with the pre-stored garment modification knowledge base. When detecting that the gap distribution numerical value of a specific body region in the structured fit data continuously exceeds the pre-set threshold, the rule engine automatically retrieves the pattern adjustment scheme corresponding to the body region from the garment modification knowledge base, and incorporates the pattern adjustment scheme as part of the specific improvement suggestion into the preliminary fit report.

[0120] Through the above steps, a preliminary fit report containing the comprehensive fit score and specific improvement suggestions is generated, providing a content basis for the visual rendering of step S43.

[0121] S43: Visual rendering is performed on the preliminary fit report generated in step S42. During the rendering process, the comprehensive fit score is displayed in the form of stars, and the score range is converted into a corresponding number of star icons through a pre-set visual mapping rule, so that the comprehensive fit score is presented in an intuitive form.

[0122] Subsequently, the specific improvement suggestions are superimposed and fused with the key frame image of the user's dynamic pose. The key frame image selects the time frame that can reflect the main action or key pose in the dynamic pose data, and takes the key frame image as a background layer.

[0123] To realize the visual fusion between layers, WebGL technology is used to load the comprehensive fit score, specific improvement suggestions, and key frame image into independent layers, respectively, and these layers are combined and rendered through a shader program.

[0124] The shader program generates corresponding heat map visual textures according to the gap distribution numerical value and the pressure mapping numerical value, and superimposes and renders the heat map textures on the user's body model in the form of color gradients, so that the user can intuitively view the fit state of each region of the body.

[0125] Finally, the visual fit report is output through an augmented reality interface, so that the user can view the comprehensive fit score, specific improvement suggestions, and the visual effect of the superimposition of the dynamic pose key frame and the heat map in the augmented reality environment.

[0126] The present application encompasses any substitutions, modifications, equivalent methods and schemes made on the essence and scope of the present application. In order for the public to have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can also be fully understood without the description of these details by those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits, etc. are not described in detail.

[0127] The above merely describes the preferred embodiments of the present application, and it should be pointed out that, for those skilled in the art, some improvements and refinements can be made without departing from the principles of the present application, and these improvements and refinements should also be considered as falling within the protection scope of the present application.

Claims

1. A dynamic pose based virtual fitting fit assessment method, characterized in that, The method comprises the following steps: Obtaining dynamic posture data of a user, the dynamic posture data comprising body joint positions and motion trajectories of the user when performing a preset action sequence; Based on the dynamic posture data, simulating the dynamic behavior of the clothes on the user's body to generate dynamic deformation data of the clothes, wherein the dynamic behavior comprises stretching, wrinkling and fitting changes of the clothes; According to the dynamic deformation data of the clothes, evaluating the fit of the clothes to generate fit evaluation indicators, the fit evaluation indicators comprising gap distribution and pressure mapping between the clothes and the body; According to the fit evaluation indicators, generating a visual fit report, the report comprising fit scores under dynamic postures and improvement suggestions.

2. The dynamic pose based virtual fitting fit assessment method according to claim 1, wherein, The step of obtaining dynamic posture data of a user comprises: Obtaining a preset action sequence, the preset action sequence comprising a plurality of standard action instructions for guiding the user to perform a dynamic posture, wherein the preset action sequence is provided to the user through a user interface display or voice prompt, and a signal is received to confirm the start of the user, to generate an executable preset action sequence; Based on the preset action sequence, real-time capture of a plurality of frames of motion data of the user performing the preset action sequence through a depth camera or an inertial measurement unit sensor, wherein the plurality of frames of motion data comprises three-dimensional point cloud information or acceleration and angular velocity data of the user's body in a time sequence, and the plurality of frames of motion data are outputted; From the plurality of frames of motion data, calculating body joint positions and motion trajectories of the user, wherein the body joint positions are extracted from the plurality of frames of motion data through a machine learning model, and the motion trajectories are derived by analyzing the changes of the body joint positions in the time dimension, and the body joint positions and the motion trajectories are outputted; According to the body joint positions and the motion trajectories, generating the dynamic posture data, wherein the dynamic posture data comprises a body joint position matrix and a motion trajectory vector, which are used as input for subsequent steps.

3. The dynamic pose based virtual fitting fit assessment method according to claim 2, wherein, The step of generating an executable preset action sequence comprises: selecting and combining a personalized preset action sequence from a predefined action library according to user's body type basic data or historical fitting records, wherein the body type basic data comprises height, weight and body type classification.

4. The dynamic pose based virtual fitting fit assessment method according to claim 2, wherein, The dynamic posture data further comprises: data standardization processing of the body joint positions and the motion trajectories, eliminating abnormal frame data in the execution process of the preset action sequence, and supplementing missing body joint positions to form complete and consistent dynamic posture data.

5. The dynamic pose based virtual fitting fit assessment method according to claim 2, wherein, The step of generating dynamic deformation data of the clothes comprises: Building a digital fabric physical model of the clothes, the digital fabric physical model comprising physical parameters of the fabric, such as tensile stiffness, bending stiffness and shear stiffness, and generating a digital fabric physical model based on the physical parameters; Coupling the digital fabric physical model with the dynamic posture data to calculate real-time deformation of the digital fabric physical model under the action of the dynamic posture data through a physics engine to generate preliminary deformation data of the clothes, wherein the real-time deformation comprises stretching, wrinkling and fitting changes of the clothes; The initial deformation data of the clothes is dynamically physically corrected, a non-penetration constraint relationship between the clothes and the body is corrected based on an energy conservation principle and a cloth collision detection algorithm, and dynamic deformation data of the corrected clothes is output, the dynamic deformation data including accurate stretch distribution, wrinkle depth and fit change amount.

6. The dynamic pose based virtual fitting fit assessment method according to claim 5, wherein, The digital fabric physical model of the constructed clothes includes: obtaining physical sample data of a target fabric through a fabric mechanical tester, and inputting the physical sample data into a material parameter inversion algorithm to calibrate physical parameters of tensile stiffness, bending stiffness and shear stiffness in the digital fabric physical model.

7. The dynamic pose based virtual fitting fit assessment method according to claim 5, wherein, The generation of the fit evaluation index includes: Based on the dynamic deformation data, a signed distance field between the clothes grid model and the user body grid model is calculated to generate an initial gap distribution, and an initial pressure mapping is calculated through a predefined pressure-stress conversion function according to a stress tensor in the dynamic deformation data of the clothes; The initial gap distribution and the initial pressure mapping are analyzed for spatiotemporal consistency, frame-to-frame mutations are smoothed through a time series filter, and data missing caused by occlusion is completed through a spatial interpolation algorithm to form a stable gap distribution and an accurate pressure mapping; According to the stable gap distribution and the accurate pressure mapping, a dynamic fit index, a local compression coefficient and an overall fit score are comprehensively calculated to generate a final fit evaluation index, wherein the fit evaluation index includes quantized gap distribution values and pressure mapping values.

8. The dynamic pose based virtual fitting fit assessment method according to claim 7, wherein, The calculation of the signed distance field between the clothes grid model and the user body grid model includes: establishing a spatial distance query data structure based on the user body grid model, and traversing the shortest directed distance from each vertex of the clothes grid model to the user body grid model, wherein a positive value represents a gap and a negative value represents penetration, thereby generating the initial gap distribution.

9. The dynamic pose based virtual fitting fit assessment method according to claim 7, wherein, The generation of the visual fit report includes: The obtained fit evaluation index is parsed, and the dynamic fit index, the local compression coefficient, the overall fit score, the gap distribution values and the pressure mapping values in the fit evaluation index are structurally processed according to a predefined report template to generate structured fit data; Based on the structured fit data, a fit report draft containing a comprehensive fit score and specific improvement suggestions is automatically generated, wherein the comprehensive fit score is calculated by weighting and fusing each value in the structured fit data, and the specific improvement suggestions are generated by matching abnormal gap distribution values and abnormal pressure mapping values in the structured fit data with a pre-stored garment modification knowledge base through a rule engine; The fit report draft is visually rendered, the comprehensive fit score is marked in the form of stars, and the specific improvement suggestions are superimposed and fused with key frame images of the user's dynamic poses, and the final visual fit report is output through an augmented reality interface.

10. The dynamic pose based virtual fitting fit assessment method according to claim 9, wherein, The structuring of the fit assessment indicators according to the predefined report template comprises: data integrity verification and outlier filtering on gap distribution values and pressure mapping values in the fit assessment indicators, and storing the verified data in corresponding data structures according to body region partitioning, to generate complete and standardized structured fit data.

Images