Garment three-dimensional model generation method and system based on artificial intelligence

Abstract

The invention discloses a garment three-dimensional model generation method and system based on artificial intelligence, and relates to the technical field of garment design, and the method comprises the steps: obtaining a human body model of a corresponding target object according to a human body image; collecting demand information of the target object on the clothes; according to the requirement for each structural feature of the garment, basic sample garments with the largest number of structural features meeting the requirement are searched in a preset sample garment database, and supplementary sample garments continue to be searched in the preset sample garment database according to the remaining structural features of the garment; stitching the supplementary sample garment and the basic sample garment according to a preset rule to obtain a target sample garment, performing size matching on the target sample garment according to the human body model, and endowing the target sample garment with corresponding attribute features to obtain a corresponding garment three-dimensional model. According to the invention, the problem of poor effect of the garment three-dimensional model generated in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of clothing design, and particularly relates to a clothing three-dimensional model generation method and system based on artificial intelligence. BACKGROUND

[0002] With the rapid development of computer graphics and artificial intelligence technology, clothing three-dimensional model generation technology has been widely applied in the fields of clothing design, virtual fitting, e-commerce display, etc.

[0003] Meanwhile, in recent years, data-driven clothing three-dimensional model generation technology has gradually emerged, such as automatically generating a three-dimensional model from a two-dimensional design drawing by using artificial intelligence technology. However, this approach still has many problems. On the one hand, it is not accurate enough in dealing with the fitting relationship between clothing and the human body, and often cannot fully consider different human body characteristics, resulting in a large difference between the wearing effect of the generated clothing model and the actual situation, affecting user experience; on the other hand, it is difficult to meet the user's demand for clothing individualization. Although the clothing model can be generated, the user's diversified demand for clothing structure (such as collar style, pocket position) and attributes (such as material) cannot be effectively integrated into the model generation process, resulting in poor effect of the generated clothing three-dimensional model. SUMMARY Therefore, the present application aims to provide a clothing three-dimensional model generation method and system based on artificial intelligence, which aims to solve the problem of poor effect of the clothing three-dimensional model generated by the prior art.

[0004] The embodiment of the present application is implemented as follows: A clothing three-dimensional model generation method based on artificial intelligence, the method comprising: obtaining a human body image containing human body information of a target object, and obtaining a human body model of the target object corresponding to the human body image by using a preset three-dimensional reconstruction algorithm according to the human body image; collecting demand information of the target object for clothing, wherein the demand information at least includes demand for structural features and attribute features of the clothing; finding out a basic sample clothing with the most structural features meeting the demand from a preset sample clothing database according to the demand for each structural feature of the clothing, and continuously finding out a supplementary sample clothing with the most structural features meeting the demand from the preset sample clothing database according to the remaining structural features of the clothing until the finding of all the structural features is completed; using the basic sample clothing as a substrate, stitching the supplementary sample clothing with the basic sample clothing according to a preset rule to obtain a target sample clothing, and giving the corresponding attribute features to the target sample clothing after size matching of the target sample clothing according to the human body model to obtain a corresponding clothing three-dimensional model.

[0005] Further, the above-mentioned garment three-dimensional model generation method based on artificial intelligence, wherein the step of finding the basic sample garment with the most structure features meeting the requirements from the preset sample garment database according to the requirements for each structure feature of the garment comprises: converting the requirements for each structure feature of the garment into corresponding structured data to obtain the structured features of the structure feature requirements of the garment; traversing the sample garment database to obtain the matching number of structure features and structured features of each sample garment, thereby finding the basic sample garment with the most structure features meeting the requirements.

[0006] Further, the above-mentioned garment three-dimensional model generation method based on artificial intelligence, wherein the step of finding the basic sample garment with the most structure features meeting the requirements from the preset sample garment database according to the requirements for each structure feature of the garment comprises: converting the three-dimensional models of the basic sample garment and the supplementary sample garment into the same format and aligning the coordinate systems to ensure that the position references of the basic sample garment and the supplementary sample garment in space are consistent; identifying the edge contours of the basic sample garment and the supplementary sample garment, and extracting feature points from the edge contours of the basic sample garment and the supplementary sample garment using a three-dimensional feature point extraction algorithm; calculating the similarity between the feature points of the edge contours of the basic sample garment and the supplementary sample garment according to the descriptors of the feature points using a feature matching algorithm, and finding the matching feature point pairs; establishing a correspondence relationship between the edge contours of the basic sample garment and the supplementary sample garment based on the matching feature point pairs, and aligning the edge contours of the basic sample garment and the supplementary sample garment; dividing the feature regions on the supplementary sample garment according to the structure features to be transferred using semantic segmentation or manual annotation, and determining the specific position and range of each feature region in the supplementary sample garment; establishing a mapping relationship between the feature regions of the supplementary sample garment and the corresponding positions of the basic sample garment based on the aligned edge contours, converting the coordinate information of the feature regions of the supplementary sample garment to the coordinate system of the basic sample garment by calculating the relative positions of the points in the feature regions and the edge contours, and determining the accurate positions of the feature regions on the substrate; combining prior knowledge of garment structure to artificially or algorithmically optimize and adjust the accurate positions of the mapped feature regions on the substrate, and then replacing the structure features on the supplementary sample garment to the basic sample garment based on the accurate positions of the feature regions on the substrate to obtain the target sample garment.

[0007] Further, the above-mentioned method for generating a three-dimensional model of a garment based on artificial intelligence, wherein the step of aligning the edge contour of the base sample garment with the edge contour of the supplementary sample garment comprises: Based on the matched feature points, the optimal rotation and translation parameters are calculated by minimizing the distance error between the matched feature points; The edge contour of the base sample garment is overlapped with the edge contour of the supplementary sample garment in space position by the optimal rotation and translation parameters, so as to align the edge contour of the supplementary sample garment with the corresponding edge contour of the base sample garment; The fine-tuning is performed by using a preset non-rigid deformation algorithm, and the edge contour of the supplementary sample garment is fitted to the base sample garment by local deformation to realize the final edge contour alignment.

[0008] Further, the above-mentioned method for generating a three-dimensional model of a garment based on artificial intelligence, wherein the step of aligning the edge contour of the base sample garment with the edge contour of the supplementary sample garment comprises: Constructing a garment adjustment model based on reinforcement learning; The accurate position of the mapped feature region on the base plate, the prior knowledge parameters of the garment structure, and the key geometric indicators of the current base sample garment are input into the garment adjustment model to obtain the accurate position coordinates of the feature region on the base plate and the updated shape parameters; The training process of the garment adjustment model based on reinforcement learning comprises: Building a network architecture based on the proximal policy optimization algorithm; Quantitative encoding of the position coordinates of the mapped feature region on the base sample garment, the shape parameters, and the prior knowledge parameters of the garment structure to construct a state space; Defining an executable action set, which includes position adjustment, shape transformation, and structure parameter modification of the feature region, and discretizing or continuous processing each action; According to the design reward function corresponding to the design target; Encoding the prior knowledge of the garment structure as an initial strategy for training as the initial input of the network architecture based on the proximal policy optimization algorithm; In each training iteration, an initial state is randomly selected from the state space as the starting state, and an action is selected from the action space according to the current strategy to perform corresponding adjustment on the feature region to obtain a new state; During the training process, the feature region is adjusted based on the current strategy each time, and the corresponding reward value is calculated, and the garment adjustment model is obtained when the average reward value calculated based on the reward value tends to be stable or reaches a preset threshold after a preset number of training rounds.

[0009] Further, the above-mentioned garment three-dimensional model generation method based on artificial intelligence, wherein the step of constructing the garment adjustment model based on reinforcement learning further comprises: receiving feedback information adjusted according to the accurate position of the feature area on the substrate, and dynamically adjusting the weight of each design target in the reward function according to the feedback information to fine-tune the garment adjustment model to obtain a final garment adjustment model.

[0010] Further, the above-mentioned garment three-dimensional model generation method based on artificial intelligence, wherein the step of assigning corresponding attribute features to the target sample garment after size matching according to the human body model to obtain a corresponding garment three-dimensional model comprises: establishing a mapping relationship between the human body model and the target sample garment, and using a three-dimensional deformation algorithm to adjust the size of the three-dimensional model of the target sample garment in real time based on the established mapping relationship, so that the size of each part of the target sample garment fits the human body model; assigning attribute features to the three-dimensional model of the matched target sample garment; checking the penetration of the human body model and the target sample garment by using a collision detection algorithm, and performing local fine-tuning on the part that collides; using a rendering engine to render the target sample garment to output a final garment three-dimensional model.

[0011] Another object of the present application is to provide a garment three-dimensional model generation system based on artificial intelligence, which comprises: a building module for obtaining a human body image containing human body information of a target object, and obtaining a human body model of the target object by using a preset three-dimensional reconstruction algorithm according to the human body image; a collection module for collecting demand information of the target object for the garment, wherein the demand information at least includes the demand for structural features and attribute features of the garment; a searching module for searching for a basic sample garment with the most structural features meeting the demand from the preset sample garment database according to the demand for each structural feature of the garment, and continuously searching for a supplementary sample garment with the most structural features meeting the demand from the preset sample garment database according to the remaining structural features of the garment until all the structural features are searched for; a generation module for taking the basic sample garment as a substrate, stitching the supplementary sample garment with the basic sample garment according to a preset rule to obtain a target sample garment, and assigning corresponding attribute features to the target sample garment after size matching according to the human body model to obtain a corresponding garment three-dimensional model.

[0012] Another object of the present application is to provide a readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to implement the steps of the method of any one of the above-mentioned methods.

[0013] It is another object of the present application to provide an electronic device comprising a memory, a processor and a computer program stored on the memory and running on the processor, said processor implementing the steps of the method described above when executing said program.

[0014] The present application acquires a human body image containing human body information of a target object, accurately constructs a human body model of the target object through a preset three-dimensional reconstruction algorithm, ensures that a garment model can closely fit a real human body shape, collects demand information of a user on garment structure features and attribute features, intelligently selects a basic sample garment and a supplementary sample garment in a sample garment database based on this, maximally matches requirements of the user on garment structure, stitches the supplementary sample garment and the basic sample garment according to a preset rule to obtain a target sample garment, and performs size matching according to the human body model, gives corresponding attribute features, and generates a high-precision garment three-dimensional model. The fitting degree of the generated garment three-dimensional model with the human body of the user is improved, and diversified demands of the user are met. The problem of poor effect of a garment three-dimensional model generated by the prior art is solved. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A flowchart of a garment three-dimensional model generation method based on artificial intelligence provided by an embodiment of the present application is shown in the figure. Figure 2 A structure block diagram of a garment three-dimensional model generation system based on artificial intelligence in a third embodiment of the present application is shown in the figure.

[0016] The following specific implementation will further illustrate the present application in combination with the above-mentioned figures. DETAILED DESCRIPTION

[0017] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the accompanying drawings. Several embodiments of the present application are shown in the drawings. However, the present application can be realized in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0018] It should be noted that when an element is referred to as being "fixedly attached" to another element, it can be directly on the other element or there can be an intervening element. When an element is referred to as being "connected" to another element, it can be directly connected to the other element or intervening elements can be present. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description herein is for describing particular embodiments only and is not intended to be limiting of the application. The use herein of the terms "and / or" includes a set of one or more associated listed items.

[0020] Embodiment one Please refer to Figure 1 , which shows the method for generating a three-dimensional clothing model based on artificial intelligence in the first embodiment of the application, which comprises steps S10-S13.

[0021] Step S10, obtain a human body image containing human body information of a target object, and obtain a human body model of the target object according to the human body image using a preset three-dimensional reconstruction algorithm.

[0022] First, professional equipment such as high-precision 3D human body scanners, depth cameras, or the cameras of terminal devices such as smartphones and tablets are used to take multiple-angle photos of the target object (i.e., the object for which a three-dimensional clothing model is to be generated), so as to obtain a human body image containing human body information of the target object. The images should cover key information such as the contour, body shape curve, and limb shape of the human body, to ensure the completeness and accuracy of the human body features. After the images are obtained, a preset three-dimensional reconstruction algorithm is used to process them, and finally a three-dimensional human body model highly matched with the body shape features of the target object is generated, providing a reliable and accurate basic model for subsequent accurate design of a three-dimensional clothing model based on the body shape, size matching, etc., effectively avoiding clothing modeling deviations caused by inaccurate human body models, and improving the overall quality and practicality of the generation of a three-dimensional clothing model. For example, the preset three-dimensional reconstruction algorithm can be based on computer vision and graphics principles, such as a three-dimensional reconstruction algorithm based on stereo vision, a three-dimensional reconstruction algorithm based on deep learning, etc. The algorithm based on stereo vision calculates the three-dimensional coordinates of each part of the human body by analyzing the parallax information in multi-view images, and then constructs a three-dimensional model; the algorithm based on deep learning trains a large number of human body images and corresponding three-dimensional model data, so that the model can learn the mapping relationship between the human body images and the three-dimensional structure, and thus directly predict a three-dimensional human body model from a single or multiple images.

[0023] Step S11, collect the demand information of the target object for clothing, wherein the demand information at least includes the demand for the structural features and attribute features of the clothing.

[0024] Specifically, the demand information at least includes the structural features and attribute features of the garment. In specific implementation, the target object can be interacted through online questionnaire, mobile application, offline communication and the like. In the information collection interface or communication content design, detailed classification and guidance are carried out around the structural features and attribute features of the garment.

[0025] For the structural features of the garment, rich options are provided, including collar type (such as lapel, stand collar, V-neck, etc.), sleeve type (such as short sleeve, long sleeve, bat sleeve, etc.), pocket design (such as patch pocket, insert pocket, three-dimensional pocket, etc.), and door style (such as zipper, button, snap, etc.). The target object can select, multiple select or freely describe according to his / her own preference, wearing scene and functional requirement, so as to comprehensively obtain the specific requirements of the target object for the structural form and combination mode of each part of the garment. In the aspect of attribute feature collection, the target object can be guided to clearly express the expectations for the attribute features of the garment, such as material (such as the preference for the texture, air permeability and durability of different fabrics such as cotton, hemp, silk and chemical fiber), color (such as the selection of basic color system and popular color, color matching and pattern requirement), and process (such as embroidery, printing, washing and wrinkle treatment). Meanwhile, the natural language processing technology can be used to analyze and classify the demand content described by the target object, and the data collected by the structured options can be integrated to form a complete and accurate garment demand information dataset. This provides detailed and accurate data basis for subsequent selection of matching garment samples in the sample garment database based on the demand information, and generation of a garment three-dimensional model meeting the individual demand of the target object, so as to ensure that the finally generated garment three-dimensional model can highly meet the actual demand of the target object in structure and attribute.

[0026] In step S12, the basic sample garment with the most satisfied structural features is found in the preset sample garment database according to the demand for each structural feature of the garment, and the supplementary sample garment with the most satisfied structural features is continuously found in the preset sample garment database according to the remaining structural features of the garment until all the structural features are found.

[0027] Wherein, after obtaining the user's demand for the structural features of the garment, the preset sample garment database can be traversed, which stores a large number of garment samples of different styles and designs, and each sample records the structural feature data of itself in detail. In the traversal process, the system will compare the structural features of each sample garment with the user's demand for the structural features one by one, and calculate the number of matching structural features of each sample garment with the user's demand according to the established matching algorithm (such as exact matching, fuzzy matching, etc.), and select the sample garment with the largest number of matching structural features as the basic sample garment, which meets the user's main demand for the structure of the garment to the greatest extent. After determining the basic sample garment, analyze the structural features in the user's demand that have not been met to form a remaining feature list, and traverse the sample garment database again to find the supplementary sample garment that meets the remaining structural features with the largest number of matching structural features. Through the continuous repetition of this process, all structural feature gaps in the user's demand are gradually filled, and all the structural feature demands put forward by the user are met, so as to complete the complete selection from the basic sample to multiple supplementary samples.

[0028] Specifically, the step of finding the basic sample garment with the largest number of matching structural features according to the demand for each structural feature of the garment in the preset sample garment database comprises: Converting the demand for each structural feature of the garment into corresponding structured data to obtain the structured features of the demand for the structural features of the garment. Traverse the sample garment database to obtain the number of matching structural features of each sample garment, so as to find the basic sample garment with the largest number of matching structural features.

[0029] In specific implementation, the user inputted clothing structure feature requirements can be standardized, that is, through specific data conversion rules and coding methods, the user's diversified and unstructured description (such as the natural language expression "want a coat with a large collar, diagonal pocket, and loose version") is converted into structured data that can be recognized and processed by a computer. Finally, a structured feature vector composed of multiple feature parameters and corresponding values is formed. After completing the structured conversion of the requirements, the preset sample clothing database is traversed, which stores a large number of clothing sample data of different styles and different structural designs, and each sample clothing also records its collar type, sleeve type, version, pocket style and other structural features in the form of structured data. In the traversal process, the system will compare the structural features of each sample clothing with the user's structured features one by one, and through matching calculation and comparison of all sample clothes in the database, the sample clothing with the highest matching number is finally selected as the basic sample clothing, which maximizes the user's requirements for clothing structural features, lays a solid foundation for subsequent selection of supplementary sample clothing and generation of the final target sample clothing, and ensures that the clothing three-dimensional model can highly fit the user's expectations in structural design, significantly improving the pertinence and practicality of clothing modeling.

[0030] In step S13, the basic sample clothing is taken as a substrate, and the supplementary sample clothing is stitched with the basic sample clothing according to the preset rules to obtain a target sample clothing, and the corresponding attribute features are given to the target sample clothing after size matching according to the human body model to obtain a corresponding clothing three-dimensional model.

[0031] In specific implementation, the selected basic sample clothing is taken as the core substrate, which has maximized the user's main requirements for clothing structure. Then, according to the pre-set stitching rules (which include professional knowledge of clothing production, such as structural compatibility, proportionality, matching of physical properties at the joint, etc.), the part of the supplementary sample clothing that meets the remaining structural feature requirements is accurately integrated into the basic sample clothing. After stitching, a target sample clothing containing all the user's structural feature requirements is generated.

[0032] Then, the system performs size matching between the target sample clothing and the human body model constructed in the early stage, and performs stretching, shrinking or local adjustment on the target sample clothing according to the accurate size data of the human body model, to ensure that the fit of each part of the clothing to the human body reaches the professional standard.

[0033] Finally, the size-matched garment model is endowed with corresponding attribute features, which come from the user demand collection stage settings, including material properties (such as hardness, elasticity), color parameters, and other physical properties (such as friction coefficient, air permeability simulation, thickness). Finally, a three-dimensional garment model is generated that not only fully meets the user's needs in structure, but also highly realistic in appearance and physical properties.

[0034] Specifically, the step of obtaining a corresponding garment three-dimensional model by assigning corresponding attribute features to the target sample garment after size matching according to the human body model comprises: Establishing a mapping relationship between the human body model and the target sample garment, based on the established mapping relationship, using a three-dimensional deformation algorithm to adjust the size of the three-dimensional model of the target sample garment in real time, so that the size of each part of the target sample garment fits the human body model; Assigning attribute features to the three-dimensional model of the matched target sample garment; Using a collision detection algorithm to check the penetration of the human body model and the target sample garment, and making local adjustments to the parts that collide; Using a rendering engine to render the target sample garment and output the final garment three-dimensional model.

[0035] In specific implementation, first, the mapping relationship between the human body model and the target sample garment is established, which requires identifying and associating the key feature points of the human body model (such as shoulder points, chest points, waist points, etc.) and the corresponding parts of the garment model (such as collar, sleeve, waistline, etc.), forming a point-to-point, region-to-region correspondence relationship, providing a positioning basis for subsequent size adjustment. Based on the established mapping relationship, a three-dimensional deformation algorithm (such as physics-based deformation, parameterized deformation, or machine learning driven deformation technology) can be used to adjust the three-dimensional model of the target sample garment in real time. These algorithms will stretch, shrink or locally deform each part of the garment according to the accurate size data of the human body model (such as chest circumference, waist circumference, height, shoulder width, etc.), ensuring that the geometric parameters such as length, circumference, and curvature of the garment are highly consistent with the human body model, for example, matching the chest circumference of a shirt with the human chest circumference, and the length of trousers with the leg length, while maintaining the overall style and structural integrity of the garment.

[0036] After the size matching is completed, the system attributes the three-dimensional model of the target sample garment with property features, which are derived from the settings in the user demand collection stage. To ensure the rationality of the garment-human interaction, the system uses collision detection algorithms (such as the GJK algorithm, bounding box detection, etc.) to check the penetration between the human model and the garment. When a collision (such as the overlapping area between the garment and the human body) is detected, the system will make local adjustments to the colliding parts, eliminating unreasonable penetration through geometric offset, surface deformation, etc., while maintaining the overall shape of the garment. Finally, the system uses professional rendering engines (such as Renderman, Arnold, or GPU-accelerated real-time rendering engines) to render the processed target sample garment. By calculating light ray tracing, shadow effects, material reflection, and other physical phenomena, a highly realistic three-dimensional garment model is generated, ultimately outputting a garment three-dimensional model that meets user needs, perfectly fits the human body, and has realistic visual effects.

[0037] In summary, the garment three-dimensional model generation method based on artificial intelligence in the above embodiments of the present application, by acquiring a human body image containing human body information of a target object, accurately constructing a human body model of the target object through a pre-set three-dimensional reconstruction algorithm, ensuring that the garment model can closely fit the real human body shape, collecting user demand information for garment structure features and property features, intelligently selecting basic sample garments and supplementary sample garments in the sample garment database based on this, maximizing the matching of user requirements for garment structure, stitching the supplementary sample garments with the basic sample garments according to the pre-set rules to obtain the target sample garment, and performing size matching according to the human body model, attributing the corresponding property features, and generating a high-precision garment three-dimensional model. The generated garment three-dimensional model is improved in terms of its adaptability to the user's body and meets the user's diverse needs. The problem of poor effect of the garment three-dimensional model generated by the prior art is solved.

[0038] Embodiment Two The garment three-dimensional model generation method based on artificial intelligence in this embodiment is different from the garment three-dimensional model generation method based on artificial intelligence in Embodiment One in that: The step of stitching the basic sample garment and the supplementary sample garment according to the pre-set rules to obtain the target sample garment includes: Converting the three-dimensional models of the basic sample garment and the supplementary sample garment into the same format and aligning the coordinate systems to ensure that the position references of the basic sample garment and the supplementary sample garment in space are consistent; Identifying the edge contours of the basic sample garment and the supplementary sample garment, and extracting feature points from the edge contours of the basic sample garment and the supplementary sample garment using a three-dimensional feature point extraction algorithm; The feature matching algorithm is used to calculate the similarity between the feature points of the edge contours of the basic sample garment and the supplementary sample garment according to the descriptors of the feature points, and the matching feature point pairs are found out; Based on the matching feature point pairs, the correspondence between the edge contours of the basic sample garment and the supplementary sample garment is preliminarily established, and the edge contours of the basic sample garment and the supplementary sample garment are aligned; On the supplementary sample garment, the feature regions are divided according to the structural features to be transferred, using semantic segmentation or manual annotation, to determine the specific position and range of each feature region in the supplementary sample garment; Taking the aligned edge contours as the reference, the mapping relationship between the feature regions of the supplementary sample garment and the corresponding positions of the basic sample garment is established, the coordinate information of the feature regions of the supplementary sample garment is converted to the coordinate system of the basic sample garment by calculating the relative positions of the points in the feature regions and the edge contours, and the accurate positions of the feature regions on the base plate are determined; Combined with the prior knowledge of garment structure, the accurate positions of the mapped feature regions on the base plate are artificially or algorithmically optimized and adjusted, and then the structural features on the supplementary sample garment are replaced onto the basic sample garment based on the accurate positions of the feature regions on the base plate to obtain the target sample garment.

[0039] Among them, the three-dimensional models of the basic sample garment and the supplementary sample garment are processed in a unified format and aligned in the coordinate system. Since three-dimensional models from different sources may use different data formats (such as OBJ, FBX, STL, etc.), and there are differences in the definition of their coordinate system origin and axis direction, format conversion can make the model data structure consistent, and coordinate system alignment ensures that the position references of the two in space are the same, laying the foundation for subsequent accurate operations.

[0040] After the format and coordinate system processing is completed, computer vision and three-dimensional modeling related algorithms are used to identify the edge contours of the basic sample garment and the supplementary sample garment, which outline the basic shape and boundary of the garment. Then, three-dimensional feature point extraction algorithms (such as Harris3D, SIFT3D, etc.) are used to extract representative feature points from the edge contours of the basic sample garment and the supplementary sample garment, which can reflect the key geometric information and local features of the contours. Then, using feature matching algorithms (such as descriptor-based matching algorithms, common descriptors include FPFH, PFH, etc.), the similarity between the feature points of the edge contours of the basic sample garment and the supplementary sample garment is calculated according to the descriptor information of the feature points, so as to find out the matching feature point pairs, which will serve as an important basis for establishing the correspondence between the edge contours of the two.

[0041] Based on the matched feature point pairs, the system preliminarily establishes the correspondence of the edge contours of the basic sample garment and the supplementary sample garment, and accurately aligns the edge contours of the two, so that the garment edges accurately correspond in space, ensuring the accuracy of subsequent splicing. On the supplementary sample garment, according to the user's demand for garment structure features, the feature areas that need to be transferred to the basic sample garment are divided by a semantic segmentation algorithm (using a deep learning model such as U-Net, DeepLab, etc. to automatically identify and segment specific structure areas) or a manual annotation method, and the specific position and range of each feature area in the supplementary sample garment are clearly defined.

[0042] Taking the aligned edge contours as the reference, the system establishes the mapping relationship between the feature areas of the supplementary sample garment and the corresponding positions of the basic sample garment, and converts the coordinate information of the feature areas of the supplementary sample garment to the coordinate system of the basic sample garment by calculating the relative positions of the points in the feature areas and the edge contours, so as to accurately determine the accurate position of the feature areas on the base plate (basic sample garment). Finally, combined with the prior knowledge of garment structure (such as garment cutting process, structure proportion specification, etc.), the position of the mapped feature areas on the base plate is adjusted through manual intervention or algorithm optimization (such as optimization algorithm based on reinforcement learning), to ensure that it meets the requirements of garment production principles and aesthetics. After optimization, the structural features on the supplementary sample garment are accurately replaced on the basic sample garment, and the target sample garment that meets the user's demand is finally obtained, providing a core foundation for subsequent generation of personalized garment three-dimensional models.

[0043] In addition, in some optional embodiments of the present application, the step of aligning the edge contours of the basic sample garment and the supplementary sample garment includes: Based on the matched feature point pairs, the optimal rotation and translation parameters are calculated by minimizing the distance error between the matched feature points; The edge contours of the basic sample garment and the supplementary sample garment are overlapped in space position by the optimal rotation and translation parameters, so as to align the edge contours of the supplementary sample garment with the corresponding edge contours of the basic sample garment; A preset non-rigid deformation algorithm is used for fine tuning, and the edge contours of the supplementary sample garment are fitted to the basic sample garment by local deformation to realize the final alignment of the edge contours.

[0044] Wherein, on the basis of matching the feature point pairs of the edge contours of the basic sample garment and the supplementary sample garment with each other, the optimal rotation and translation parameters are solved by iterative calculation in the form of minimizing the distance error between the matching feature points. Specifically, this process will consider the X, Y and Z axis directions in the three-dimensional space, and by constantly adjusting the rotation angle and displacement distance, the spatial distance between each group of matching feature points is reduced as much as possible, and finally the rotation and translation parameter combination that can make the two groups of edge contours reach the best corresponding state in the spatial position is obtained.

[0045] After obtaining the optimal rotation and translation parameters, the parameters are applied to the edge contour of the supplementary sample garment, and the corresponding rotation and translation transformation is performed on the supplementary sample garment, so that the edge contour corresponding to the basic sample garment in the spatial position is overlapped. This step preliminarily completes the alignment of the edge contours of the two, so that the edges of the supplementary sample garment and the basic sample garment have consistency in the spatial layout, but due to the complexity and diversity of the shape of the garment, only rigid rotation and translation may not be able to achieve complete fitting.

[0046] Therefore, a preset non-rigid deformation algorithm is used to further fine-tune the edge contour of the supplementary sample garment. The non-rigid deformation algorithm can perform local deformation operation on the garment model, which can stretch, shrink or bend the local area of the edge contour of the supplementary sample garment according to the shape and curvature change of the edge contour of the basic sample garment, so that the edge contour of the supplementary sample garment can closely fit the edge contour of the basic sample garment, eliminate the gap or misalignment caused by shape difference, and thus realize the final precise and natural edge contour alignment effect.

[0047] In addition, in some optional embodiments of the present application, the step of combining the prior knowledge of the garment structure to artificially or algorithmically optimize and adjust the accurate position of the mapped feature area on the substrate includes: constructing a garment adjustment model based on reinforcement learning; inputting the accurate position of the mapped feature area on the substrate, the prior knowledge parameters of the garment structure and the key geometric indicators of the current basic sample garment into the garment adjustment model to obtain the accurate position coordinates of the feature area on the substrate and the updated shape parameters; Wherein, the training process of the garment adjustment model based on reinforcement learning includes: building a network architecture based on the proximal policy optimization algorithm; quantitative encoding the position coordinates of the mapped feature area on the basic sample garment, the shape parameters, and the prior knowledge parameters of the garment structure to construct a state space; define a set of executable actions, the action set includes position adjustment, shape transformation, structure parameter modification of the feature region, discretize or continuous each action; a design reward function corresponding to the design target; Encode the prior knowledge of the garment structure as an initial strategy, which is the initial input of the network architecture based on the proximal policy optimization algorithm for training; In each training iteration, an initial state is randomly selected from the state space as the starting state, and an action is selected from the action space according to the current policy to execute, and the corresponding adjustment of the feature region is obtained. New state; During the training process, the feature region is adjusted based on the current policy each time, and the corresponding reward value is calculated, and the average reward value calculated based on the reward value tends to be stable or reaches a preset threshold after a preset number of training. Obtain the garment adjustment model.

[0048] Among them, by constructing a garment adjustment model based on reinforcement learning, more accurate optimization of garment structure features is realized. Specifically, first, build a network architecture based on the proximal policy optimization algorithm (PPO). PPO algorithm as an efficient policy optimization algorithm in the field of reinforcement learning, can effectively learn and optimize decision-making strategy in complex environment, provide core algorithm support for garment adjustment model; Subsequently, the position coordinates of the mapped feature region on the basic sample garment, the shape parameters, and the prior knowledge parameters of the garment structure (such as the reasonable range of collar angle, the standard value of sleeve pocket depth, etc.) are quantitatively coded. These information is integrated and constructed as the state space of the model. Each state is presented in the form of a multi-dimensional vector, fully reflecting the current feature region layout and garment structure constraint conditions, providing decision basis for the model. At the same time, define a set of executable actions, covering position adjustment (translation along X, Y, Z axis by a certain distance), shape transformation (scaling adjustment, angle rotation), structure parameter modification (collar width increase or decrease, pocket position movement, etc.) operations on the feature region, and discretize or continuous each action to meet the algorithm calculation requirements, forming the action space of the model; To guide the model learning direction, according to the design target (such as beauty, practicality, conformity with popular trends, etc.), the corresponding design reward function is formulated. For example, the beauty reward can be based on comparison with a fashion style database to calculate the feature similarity between the garment and similar style samples; the practicality reward can be measured by simulating the stress distribution uniformity of each part of the garment during human motion; if there is a structure conflict or violates the prior knowledge constraint, it will be punished. Encode the prior knowledge of the garment structure as an initial strategy, input the network architecture based on PPO algorithm as the training starting point, which contains industry standards, common design rules and other knowledge, and provides a basic direction for model optimization; In the training process, an initial state is randomly selected from the state space each time iteration, and the system selects an action from the action space according to the current policy to execute, and obtains a new state after adjusting the feature area accordingly. After each adjustment, the reward value is calculated according to the designed reward function to evaluate the decision-making effect of the current policy. As the training progresses, the model continuously accumulates data such as states, actions, and rewards, and gradually adjusts the policy parameters through the policy gradient optimization and value network update mechanism of the PPO algorithm, so that the policy is optimized in the direction of obtaining higher rewards. When the training reaches the preset number of rounds, the average reward value calculated according to the reward value is continuously observed. If it tends to be stable or reaches a preset threshold, it indicates that the model has converged and obtained a stable and effective optimized policy. Finally, a garment adjustment model that can accurately optimize the position and shape parameters of the feature area is obtained. Subsequently, the accurate position of the mapped feature area on the base plate, the prior knowledge parameters of the garment structure, and the key geometric indicators (such as the size and proportion of each part) of the current basic sample garment are input into the trained model, which can output the accurate position coordinates of the optimized feature area on the base plate and the updated shape parameters, ensuring that the structure design of the target sample garment meets the user's individual needs and follows the professional specifications of garment structure, significantly improving the quality and practicality of the generated garment three-dimensional model.

[0049] In addition, in some optional embodiments of the present application, the step of constructing a garment adjustment model based on reinforcement learning further comprises: receiving feedback information for the accurate position adjustment of the feature area on the base plate, and dynamically adjusting the weights of each design target in the reward function according to the feedback information to fine-tune the garment adjustment model to obtain a final garment adjustment model.

[0050] Wherein, after completing the clothing adjustment model based on reinforcement learning, in order to further improve the adaptation ability of the model to the personalized needs of the user, a feedback optimization mechanism is added. When the user or the operator observes the position adjustment result of the feature area on the substrate, feedback information can be sent to the system through the interactive interface or a specific channel. These feedback information covers the evaluation of the beauty, practicality, process rationality and other aspects of the adjusted clothing structure, for example, the user thinks that the collar of the clothing adjusted this time meets the structural specification, but still needs to be improved from the aesthetic point of view, or feels that the position adjustment of a pocket affects the practicality. After receiving the feedback information, the system will analyze it and extract the user's emphasis on different design goals (such as beauty, practicality, compliance with prior knowledge, etc.). Based on the analysis result, the system dynamically adjusts the weight of each design goal in the reward function. If the user feedback pays more attention to beauty, the weight of beauty in the reward function is increased, and the weight of other goals is reduced. Conversely, the same is true. By adjusting the weight, the importance of different design goals in the training and optimization process of the model is changed, guiding the model to fine-tune in the direction that better meets the user's expectations. In subsequent training iterations, the model re-evaluates the reward value obtained after adjusting the feature area according to the new reward function weight, and then optimizes the strategy parameters. After several iterations of fine-tuning, the final clothing adjustment model that better meets the user's personalized needs is obtained, so that the generated clothing three-dimensional model not only meets the design specifications, but also better meets the user's subjective aesthetic and actual use requirements.

[0051] In summary, the clothing three-dimensional model generation method based on artificial intelligence in the above embodiments of the present application acquires a human body image containing human body information of a target object, accurately constructs a human body model of the target object through a preset three-dimensional reconstruction algorithm, ensures that the clothing model can closely fit the real human body shape, collects user demand information for clothing structure features and attribute features, intelligently selects basic sample clothing and supplementary sample clothing in the sample clothing database based on this, maximizes the matching of user requirements for clothing structure, stitches the supplementary sample clothing and the basic sample clothing according to the preset rule to obtain a target sample clothing, and performs size matching according to the human body model, assigns corresponding attribute features, and generates a high-precision clothing three-dimensional model. The adaptability of the generated clothing three-dimensional model to the user's body is improved, and the diversified needs of the user are met. The problem of poor effect of the clothing three-dimensional model generated by the prior art is solved.

[0052] Embodiment three Please refer to Figure 2 , which shows the clothing three-dimensional model generation system based on artificial intelligence proposed in the third embodiment of the present application, the system comprises: The establishment module 100 is used for acquiring a human body image containing human body information of a target object, and obtaining a corresponding human body model of the target object by using a preset three-dimensional reconstruction algorithm according to the human body image. The acquisition module 200 is configured to acquire demand information of the target object for the clothes, wherein the demand information at least includes demands for structural features and attribute features of the clothes; The searching module 300 is configured to search for a basic sample clothes with the most structural features meeting the demands from the preset sample clothes database according to the demand for each structural feature of the clothes, and continue to search for a supplementary sample clothes with the most structural features meeting the demands from the preset sample clothes database according to the remaining structural features of the clothes until all the structural features are searched. The generating module 400 is configured to take the basic sample clothes as a substrate, sew the supplementary sample clothes with the basic sample clothes according to a preset rule to obtain a target sample clothes, and give corresponding attribute features to the target sample clothes after size matching of the target sample clothes according to the human body model to obtain a corresponding clothes three-dimensional model.

[0053] The functions or operation steps realized when the above modules are executed are generally the same as those of the above method embodiments, and thus will not be described here again.

[0054] Embodiment Four Another aspect of the present application further provides a readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to realize the steps of the method according to any one of the above embodiments one to two.

[0055] Embodiment Five Another aspect of the present application further provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to realize the steps of the method according to any one of the above embodiments one to two.

[0056] The technical features of each of the above embodiments can be combined arbitrarily, and to make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0057] Those skilled in the art will appreciate that the logic and / or steps represented in the flow diagrams, or otherwise described herein, can be embodied in

[0058] More specific examples (a non-exhaustive list) of the computer readable storage medium include the following: an electrical connection having one or more wires (electrical device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer readable storage medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

[0059] It should be understood that aspects of the application can be implemented in hardware, software, firmware, or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example in software or firmware, stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following technologies, known in the art, or combinations thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0060] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Also, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0061] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for generating 3D clothing models based on artificial intelligence, characterized in that, The method includes: Acquire human images containing human information of the target object, and obtain the corresponding human model of the target object based on the human images using a preset 3D reconstruction algorithm; Collect information on the target audience's needs for clothing, including at least their requirements for the structural and attribute features of clothing. Based on the requirements for each structural feature of the garment, the system searches the preset sample garment database for the basic sample garment that meets the requirements in terms of the number of structural features that meet the requirements. Then, based on the remaining structural features of the garment, the system continues to search the preset sample garment database for the supplementary sample garment that meets the requirements in terms of the number of structural features that meet the requirements until all structural features have been searched. Using the basic sample garment as a base, the supplementary sample garment is sewn to the basic sample garment according to preset rules to obtain the target sample garment. After matching the size of the target sample garment with the human body model, the corresponding attribute features are assigned to obtain the corresponding three-dimensional garment model.

2. The method for generating a 3D clothing model based on artificial intelligence according to claim 1, characterized in that, The step of finding the basic sample garment with the most structural features that meet the requirements of each structural feature of the garment in a pre-set sample garment database includes: The requirements for each structural feature of clothing are converted into corresponding structured data, resulting in structured features of the requirements for the structural features of clothing. Traverse the sample clothing database to obtain the number of matches between the structural features and structured features of each sample clothing, thereby finding the basic sample clothing with the most structural features that meet the requirements.

3. The method for generating a 3D clothing model based on artificial intelligence according to claim 1, characterized in that, The step of using a basic sample garment as a base and sewing a supplementary sample garment to the basic sample garment according to a preset rule to obtain the target sample garment includes: The 3D models of the basic sample garment and the supplementary sample garment are converted into the same format and their coordinate systems are aligned to ensure that the basic sample garment and the supplementary sample garment have the same spatial reference. Identify the edge contours of the basic sample clothing and the supplementary sample clothing, and extract feature points from the edge contours of the basic sample clothing and the supplementary sample clothing using a three-dimensional feature point extraction algorithm; Using a feature matching algorithm, the similarity between feature points of the edge contours of the base sample clothing and the supplementary sample clothing is calculated based on the descriptors of the feature points, and the matching feature point pairs are found. Based on the matched feature point pairs, a preliminary correspondence between the edge contours of the basic sample clothing and the supplementary sample clothing is established, and the edge contours of the basic sample clothing and the supplementary sample clothing are aligned. In the supplementary sample garments, based on the structural features to be transferred, feature regions are divided using semantic segmentation or manual annotation to determine the specific location and range of each feature region in the supplementary sample garments. Based on the aligned edge contour, a mapping relationship is established between the feature areas of the supplementary sample garment and the corresponding positions of the base sample garment. By calculating the relative positions of the points within the feature area and the edge contour, the coordinate information of the feature areas of the supplementary sample garment is transformed into the coordinate system of the base sample garment, and the accurate position of the feature areas on the substrate is determined. By combining prior knowledge of clothing structure, the accurate position of the mapped feature region on the substrate is optimized and adjusted manually or by algorithm. Then, based on the accurate position of the feature region on the substrate, the structural features of the supplementary sample clothing are replaced on the basic sample clothing to obtain the target sample clothing.

4. The method for generating a 3D clothing model based on artificial intelligence according to claim 3, characterized in that, The step of aligning the edge contours of the base sample garment and the supplementary sample garment includes: Based on the matched feature point pairs, the optimal rotation and translation parameters are calculated by minimizing the distance error between the matched feature points. By using optimal rotation and translation parameters, the edge contours of the base sample garment and the supplementary sample garment are made to coincide in spatial position, so as to align the edge contours of the supplementary sample garment with the corresponding edge contours of the base sample garment. A preset non-rigid deformation algorithm is used for fine-tuning. Through local deformation, the edge contour of the supplementary sample garment is made to fit the basic sample garment, thus achieving the final edge contour alignment.

5. The method for generating a 3D clothing model based on artificial intelligence according to claim 1, characterized in that, The step of manually or algorithmically optimizing the accurate position of the mapped feature region on the substrate by incorporating prior knowledge of the garment structure includes: Construct a clothing adjustment model based on reinforcement learning; The accurate position of the mapped feature region on the substrate, the prior knowledge parameters of the garment structure, and the key geometric indicators of the current basic sample garment are input into the garment adjustment model to obtain the accurate position coordinates of the feature region on the substrate and the updated shape parameters. The training process of the reinforcement learning-based clothing adjustment model includes: Construct a network architecture based on a near-end policy optimization algorithm; The mapping of the feature regions on the base sample clothing is quantized and encoded to construct the state space, including the position coordinates, shape parameters, and prior knowledge parameters of the clothing structure. Define a set of executable actions, which includes adjusting the position of the feature region, transforming its shape, and modifying its structural parameters. Each action can be discretized or processed continuously. Based on the design reward function corresponding to the design objective; Prior knowledge of clothing structure is encoded as an initial policy and used as the initial input for training the network architecture based on the proximal policy optimization algorithm. In each training iteration, an initial state is randomly selected from the state space as the starting state, and an action is selected from the action space according to the current policy to be executed. The feature regions are then adjusted accordingly to obtain a new state. During the training process, the feature region is adjusted based on the current strategy each time and the corresponding reward value is calculated. After a preset number of training rounds, the average reward value calculated based on the reward value tends to stabilize or reaches a preset threshold to obtain the clothing adjustment model.

6. The method for generating a 3D clothing model based on artificial intelligence according to claim 1, characterized in that, The step of constructing a reinforcement learning-based clothing adjustment model also includes: The system receives feedback information after adjusting the precise position of the feature region on the substrate, and dynamically adjusts the weights of each design objective in the reward function based on the feedback information to fine-tune the clothing adjustment model and obtain the final clothing adjustment model.

7. The method for generating a 3D clothing model based on artificial intelligence according to any one of claims 1 to 6, characterized in that, The steps of matching the size of the target sample clothing based on the human body model and then assigning corresponding attribute features to obtain the corresponding 3D clothing model include: Establish a mapping relationship between the human body model and the target sample clothing. Based on the established mapping relationship, use a three-dimensional deformation algorithm to adjust the size of the three-dimensional model of the target sample clothing in real time so that the size of each part of the target sample clothing fits the human body model. Assign attribute features to the 3D model of the matched target sample clothing; Collision detection algorithms are used to check the penetration between the human body model and the target sample clothing, and local fine-tuning is performed on the areas where collisions occur. The rendering engine is used to render the target sample clothing and output the final 3D model of the clothing.

8. A system for generating 3D clothing models based on artificial intelligence, characterized in that, The system includes: The module is used to acquire human images containing human information of the target object, and to obtain the corresponding human model of the target object based on the human images using a preset 3D reconstruction algorithm. The data acquisition module is used to collect the target object's demand information for clothing, which includes at least the demand for the structural features and attribute features of clothing. The search module is used to find the basic sample garment with the most structural features that meet the requirements of each structural feature of the garment in the preset sample garment database, and to continue to search for the supplementary sample garment with the most structural features that meet the requirements in the preset sample garment database based on the remaining structural features of the garment until all structural features have been searched. The generation module is used to take the basic sample garment as a base, sew the supplementary sample garment to the basic sample garment according to preset rules to obtain the target sample garment, and then assign corresponding attribute features to the target sample garment after size matching according to the human body model to obtain the corresponding three-dimensional garment model.

9. A readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the program, implements the steps of the method as described in any one of claims 1 to 7.

Images