Parametric plantar surface automatic generation method and system

Abstract

The application provides an automatic generation method and system based on a parameterized foot bottom curve, and relates to the fields of computer graphics and digitization technology. First, foot size data and style reference images of a target user and randomly initialized three-dimensional data are obtained. Second, foot mechanics data are generated based on the foot size data. Third, contour features and curvature features are extracted and converted into geometric change information. Fourth, the randomly initialized three-dimensional data are iteratively updated to obtain an initial three-dimensional foot bottom curve. Finally, in the intermediate expression space, a first type of intermediate variable is distinguished from a second type of intermediate variable, and the target parameterized foot bottom curve is obtained through modification of the first type of intermediate variable. The technical scheme provided by the application not only realizes independent and flexible adjustment of the appearance form while maintaining the stability of the support performance of the curve, but also completes efficient and personalized foot bottom curve customization design.

Description

Technical Field

[0001] This application relates to the fields of computer graphics and digital technology, and in particular to a method and system for automatically generating parametric foot surfaces. Background Technology

[0002] With the growing demand for personalized customization, there is an urgent need for a method that can quickly and automatically generate the sole surface in the digital design of footwear, orthotics, and bionic robot foot components.

[0003] Currently, the existing technology adopts a three-dimensional foot surface generation scheme based on a deep learning model. This scheme first obtains geometric data through three-dimensional scanning of the foot and extracts mechanical feature parameters from it. At the same time, the style image selected by the user is encoded with image features. Then, the mechanical parameters and style feature encoding are input into a pre-trained generative neural network. The network directly outputs a three-dimensional mesh model that integrates function and style as the design result.

[0004] However, this existing solution tightly couples mechanical properties and stylistic features at the model's underlying layer during the generation process, resulting in a black box-like 3D model. When detailed adjustments or stylistic modifications are needed to the visual appearance of the generated result, any editing operation targeting the appearance will uncontrollably alter the model's mechanical structure and properties, compromising its initial mechanical performance. This prevents designers from making independent, refined adjustments and redesigns to the aesthetic form while maintaining the stability of the product's core functions, severely limiting design flexibility and the depth of personalized customization. Summary of the Invention

[0005] This application provides a method and system for automatically generating parametric plantar surfaces, which solves the problem in the prior art that the appearance of the plantar surface cannot be independently and finely adjusted while maintaining the stability of its mechanical properties, resulting in a rigid design process and low efficiency.

[0006] Firstly, this application provides a method for automatically generating parametric plantar surfaces, including:

[0007] Acquire the target user's foot size data and style reference images, as well as a set of randomly initialized 3D data;

[0008] Based on the foot size data, foot biomechanical data is generated;

[0009] Contour features and curvature features are extracted from the style reference image, and the contour features and curvature features are converted into geometric transformation information;

[0010] Based on the foot biomechanics data and the geometric change information, the randomly initialized three-dimensional data is iteratively updated to obtain the initial three-dimensional plantar surface;

[0011] In the intermediate expression space corresponding to the initial three-dimensional plantar surface, the first type of intermediate variables that determine the appearance and the second type of intermediate variables that determine the mechanical properties are distinguished and controlled. By modifying the first type of intermediate variables, the target parameterized plantar surface is obtained.

[0012] Optionally, based on the foot size data, foot biomechanical data is generated, including:

[0013] Based on the multiple dimensional components contained in the foot size data, an initial foot biomechanical feature representation is generated;

[0014] The gait pattern identifier of the target user, the foot surface contour feature information of the target user, and the foot movement sequence information of the target user during the movement process are obtained.

[0015] The gait pattern identifier is combined with the initial foot biomechanical feature representation to form an intermediate biomechanical feature representation;

[0016] Based on the foot movement sequence information, the dominant movement amplitude of the target user's foot in multiple movement directions is calculated;

[0017] The dominant motion amplitude is superimposed on the intermediate mechanical feature expression to form an enhanced mechanical feature expression;

[0018] The foot surface contour feature information is combined with the enhanced mechanical feature expression to generate foot biomechanical data.

[0019] Optionally, contour features and curvature features are extracted from the style reference image, and the contour features and curvature features are converted into geometric transformation information, including:

[0020] Identify the target shape boundary lines in the style reference image to obtain contour feature information. The target shape boundary lines include the outer contour lines and the main internal segmentation lines of the object depicted in the style reference image.

[0021] The curvature feature information of the style reference image is obtained by measuring the curvature change at each point on the boundary line of the target shape.

[0022] Establish the feature correspondence between the style reference image and the preset standard three-dimensional foot basic shape;

[0023] Based on the feature correspondence, the morphological trend of the target shape boundary lines described by the contour feature information and the curvature change described by the curvature feature information are jointly mapped into geometric change information.

[0024] Optionally, based on the foot biomechanics data and the geometric change information, the randomly initialized three-dimensional data is iteratively updated to obtain an initial three-dimensional plantar surface, including:

[0025] The foot biomechanics data and the geometric change information are combined to form a combination of conditions to guide the generation of the curved surface;

[0026] Based on the combination of conditions and randomly initialized 3D data, calculate the update direction of the 3D data in the current iteration round;

[0027] Based on the combination of conditions, calculate the adjustment range of the 3D data along the update direction in the current iteration round;

[0028] The three-dimensional data is updated according to the update direction and the adjustment range to obtain the updated three-dimensional data;

[0029] Determine whether the updated 3D data meets the preset convergence condition. If the updated 3D data meets the convergence condition, then the updated 3D data is output as the initial 3D foot surface.

[0030] If the updated 3D data does not meet the convergence condition, the updated 3D data is used as the 3D data in the next iteration round. The steps of calculating the update direction, calculating the adjustment range, updating the 3D data, and determining whether the convergence condition is met are repeated until the convergence condition is met to output the initial 3D foot surface.

[0031] Optionally, in the intermediate representation space corresponding to the initial three-dimensional plantar surface, the first type of intermediate variables determining the appearance and the second type of intermediate variables determining the mechanical properties are distinguished and controlled. By modifying the first type of intermediate variables, the target parameterized plantar surface is obtained, including:

[0032] The initial three-dimensional plantar surface is transformed into an intermediate form to obtain the spatial state representation of the initial three-dimensional plantar surface in the intermediate expression space.

[0033] From the spatial state representation, a first subset corresponding to the first type of intermediate variables and a second subset corresponding to the second type of intermediate variables are separated.

[0034] Receive editing input information for the first type of intermediate variables;

[0035] Based on the edit input information, the intermediate variables contained in the first subset are adjusted to form the adjusted first subset;

[0036] The adjusted first subset is combined with the second subset to form an updated spatial state representation;

[0037] The updated spatial state representation is reverse-transformed to generate the target parameterized plantar surface.

[0038] Optionally, based on the combination of conditions, the adjustment magnitude of the 3D data along the update direction in the current iteration is calculated, including:

[0039] Based on the foot biomechanics data contained in the combination of conditions, determine the constraint boundaries used to limit the shape changes of the three-dimensional curved surface;

[0040] In the current iteration, identify the target surface region in the 3D data that interacts with the constraint boundary;

[0041] Based on the geometric change information contained in the combination of conditions, calculate the expected intensity of morphological adjustment for the target surface region;

[0042] Based on the current iteration round number, a dynamic adjustment coefficient related to the iteration round is generated;

[0043] Based on the relative relationship between the target surface region and the constraint boundary, the expected intensity of the morphological adjustment, and the dynamic adjustment coefficient, the adjustment magnitude of the 3D data along the update direction in the current iteration is calculated.

[0044] Optionally, separating a first subset corresponding to the first type of intermediate variables and a second subset corresponding to the second type of intermediate variables from the spatial state representation includes:

[0045] A relational network is constructed for the spatial state representation, wherein the nodes of the relational network correspond to the intermediate variables in the spatial state representation, and the edges of the relational network represent the interaction strength between the corresponding intermediate variables.

[0046] In the network of relationships, based on the plantar surface attributes described by the data content corresponding to the intermediate variables, each intermediate variable is labeled with a category label indicating whether it belongs to the first attribute or the second attribute, thus obtaining the first attribute intermediate variable and the second attribute intermediate variable.

[0047] In the network of relationships, a first sub-network region formed by interconnecting intermediate variables of the first attribute and a second sub-network region formed by interconnecting intermediate variables of the second attribute are identified.

[0048] From the spatial state representation, extract the set of intermediate variables corresponding to the nodes of the first sub-network region to form the first subset;

[0049] From the spatial state representation, the set of intermediate variables corresponding to the nodes of the second sub-network region is extracted to form the second subset.

[0050] Secondly, this application provides an automatic generation system for parametric plantar surfaces, comprising:

[0051] The acquisition module is used to acquire the target user's foot size data and style reference image, as well as a set of randomly initialized 3D data;

[0052] A generation module is used to generate foot biomechanical data based on the foot size data;

[0053] The conversion module is used to extract contour features and curvature features from the style reference image and convert the contour features and curvature features into geometric transformation information;

[0054] An update module is used to iteratively update the randomly initialized three-dimensional data based on the foot biomechanical data and the geometric change information to obtain an initial three-dimensional plantar surface;

[0055] The modification module is used to distinguish and control the first type of intermediate variables that determine the appearance and the second type of intermediate variables that determine the mechanical properties in the intermediate expression space corresponding to the initial three-dimensional plantar surface. By modifying the first type of intermediate variables, the target parameterized plantar surface is obtained.

[0056] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the automatic generation method based on parametric plantar surface as described in the first aspect above.

[0057] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a method for automatically generating parametric plantar surfaces as described in the first aspect.

[0058] This application obtains foot size and style reference images of target users, generates foot biomechanical data and geometric change information, and uses these to iteratively update randomly initialized 3D data, thereby obtaining an initial 3D plantar surface that simultaneously considers functional adaptation and style integration. Crucially, in the intermediate expression space corresponding to the initial surface, variables that determine appearance and variables that determine biomechanical properties are controlled separately. This allows the appearance of the surface to be adjusted only by modifying the first type of intermediate variables in the subsequent design stage, while the second type of variables that determine biomechanical properties remain constant. Thus, technically, it achieves independent and precise adjustment of the appearance of the plantar surface while ensuring the stability of its biomechanical properties.

[0059] Furthermore, by constructing a relational network for the state representation of the intermediate expression space and classifying the attributes of the foot surface described by the intermediate variables, it is possible to automatically identify and separate sub-network regions dominated by appearance attribute variables and mechanical attribute variables, respectively. This separation method based on the intrinsic attributes and interrelationships of variables ensures that the division between the first subset and the second subset is not arbitrary, but faithfully reflects the essential structure and coupling relationship of the two major design dimensions of form and function in the surface model, laying a reliable data foundation for subsequent accurate and non-interfering independent editing.

[0060] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 A flowchart of an automatic generation method for parametric plantar surfaces provided in this application is shown;

[0063] Figure 2 This paper presents a schematic diagram of a parametric plantar surface automatic generation system provided in this application.

[0064] Figure 3 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0065] To enable those skilled in the art to better understand the present application, the technical solution of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0066] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0067] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0068] Figure 1 This application provides a flowchart of a method for automatically generating parametric plantar surfaces, such as... Figure 1 As shown, the method includes:

[0069] Step 101: Obtain the target user's foot size data and style reference image, as well as a set of randomly initialized 3D data.

[0070] In this step, foot size data refers to a set of quantified values ​​describing the key physical dimensions of the target user's foot, used to provide basic anatomical constraints for the functional design of the foot's surface. It is usually obtained through 3D foot scanner measurements or 3D reconstruction techniques based on multi-view photographs.

[0071] Style reference images refer to any two-dimensional images provided by users that embody their desired visual style of the sole surface, such as streamlined or biomimetic forms. These images provide aesthetic guidance and morphological inspiration for surface generation and are obtained by users through selection from an image library or by directly uploading their personal images.

[0072] Randomly initialized 3D data refers to a set of coordinates of a large number of spatial points generated by a computer in 3D space according to a preset random distribution rule such as Gaussian distribution. It does not have any specific shape semantics and is used as the initial state or starting point of the entire parametric surface generation process. It is dynamically created during program initialization by calling a pseudo-random number generator.

[0073] In this step, firstly, a structured light 3D scanner is used to scan the target user's foot. This scanner projects a series of coded grating patterns onto the foot surface, and a camera captures the patterns deformed by the undulations of the foot surface. Then, using optical triangulation, the 3D coordinates of a large number of points on the foot surface are calculated, resulting in a dense point cloud dataset that accurately reflects the foot's shape, containing multiple dimensions such as length, width, and height—i.e., foot size data. Secondly, a graphical user interface is designed to allow users to browse and select one or more images from local storage or an online style library as input. These user-selected image files are received and decoded into standard RGB pixel matrices, stored in memory; these images serve as style reference images. Finally, when the software program begins the generation process, a pseudo-random number generation function is called to randomly generate hundreds of thousands of 3D coordinate points within a predefined cubic area large enough to encompass the entire foot space, either uniformly or according to a Gaussian distribution. Each point is represented by X, Y, ... The three coordinates Z and Z are used to form a set of points that are initially random and do not form any meaningful surface shape. This set of data is the randomly initialized three-dimensional data. At this point, the three types of basic input data with different properties required for the subsequent generation process are all ready.

[0074] For example, Company A's custom shoe design platform creates a foot surface design for a user. First, it guides the user to place their foot on a B-type desktop 3D scanner connected to the platform. After scanning, the platform obtains the user's precise foot point cloud data, i.e., foot size data. Second, the platform interface displays a library of images containing various styles, such as sports car curves, mountain outlines, and futuristic mecha. Then, the user selects a side view of a sports car with a streamlined outline as a source of inspiration, and this image is uploaded as a style reference image. At the same time, the platform's backend generation program automatically starts, and its initialization module instantly generates 1 million randomly distributed spatial point coordinates within the 3D spatial boundary corresponding to the foot size, forming the starting point of the randomly initialized 3D data for this design task.

[0075] Step 102: Generate foot biomechanical data based on the foot size data.

[0076] Optionally, step 102 may specifically include:

[0077] Step 1021: Generate an initial foot biomechanical feature representation based on the multiple dimensional components contained in the foot size data.

[0078] Step 1022: Obtain the gait pattern identifier of the target user, the foot surface contour feature information of the target user, and the foot movement sequence information of the target user during the movement process.

[0079] Step 1023: Combine the gait pattern identifier with the initial foot biomechanical feature expression to form an intermediate biomechanical feature expression.

[0080] Step 1024: Based on the foot movement sequence information, calculate the dominant movement amplitude of the target user's foot in multiple movement directions.

[0081] Step 1025: The dominant motion amplitude is superimposed on the intermediate mechanical feature expression to form an enhanced mechanical feature expression.

[0082] Step 1026: Combine the foot surface contour feature information with the enhanced mechanical feature expression to generate foot mechanical data.

[0083] In this step, foot biomechanical data refers to a structured set of data that comprehensively characterizes the pressure distribution, joint load, and stability of the target user's foot during movement, and is used to provide a quantitative basis for the biomechanical adaptation design of the foot surface.

[0084] Dimensional components refer to the individual measurement indicators that make up foot size data, such as foot length, foot width, arch height, and ankle circumference. They are used to describe the specific geometric features of the foot in three-dimensional space and are the original inputs for generating initial mechanical features.

[0085] Initial foot biomechanical characteristic representation refers to a mathematical vector or data structure derived from multiple dimensional components of foot size data through an algorithm, which preliminarily describes the static biomechanical properties of the foot. It is used to establish a mapping from size to basic biomechanical performance and is calculated through a specific biomechanical statistical model.

[0086] Gait pattern identifiers refer to a label or code used to classify and identify the characteristic walking patterns of a target user, such as pigeon toes or normal gait. They are used to introduce personalized dynamic behavior patterns in biomechanical analysis, either by analyzing plantar pressure maps or by the user selecting from preset options.

[0087] Foot surface contour feature information refers to the digital description of the uneven shape of the surface of the three-dimensional model of the foot, especially the curvature of the arch and the degree of protrusion of the metatarsal heads. It is used to reflect the inherent morphological characteristics of the foot when it is in contact with the supporting surface. It is obtained by surface reconstruction and geometric analysis of the three-dimensional scan point cloud of the foot.

[0088] Foot motion sequence information refers to the data stream recorded by sensors of the position and angle changes of the foot in three-dimensional space over time during a complete gait cycle of a target user. It is used to analyze the dynamic kinematic characteristics of the foot and is acquired through an inertial measurement unit (IMU).

[0089] Intermediate mechanical feature representation refers to the transitional mechanical description formed by incorporating gait pattern identifier information into the initial foot mechanical feature representation. It is used to reflect the influence of personalized gait on static mechanical features and is generated through feature splicing or conditional coding techniques.

[0090] The dominant motion amplitude refers to the typical amplitude value extracted from foot motion sequence information, which is the motion in a specific direction such as left and right flips or forward and backward rolls. It is used to quantify the dynamic range of foot activity and is obtained by principal component analysis or extreme value statistics on motion sequence data.

[0091] Enhanced mechanical feature representation refers to a more comprehensive mechanical description formed by superimposing dominant motion amplitude information on the intermediate mechanical feature representation. It is used to simultaneously cover static, gait pattern and dynamic motion range information, and is achieved through vector addition or feature space expansion operations.

[0092] In this step, foot size data is first processed through a pre-trained multilayer perceptron network. This multilayer perceptron network takes multiple dimensional components contained in the size data, such as foot length and foot width, as input features and performs a series of nonlinear weighted calculations and transformations on these input features. Finally, it outputs a numerical vector of fixed length. This numerical vector can encode the complex relationship between size and basic mechanical properties, such as the estimated position of the plantar pressure center and the stiffness coefficient of the arch. This output vector is the initial expression of foot mechanical features.

[0093] Secondly, three types of additional information are acquired through three parallel methods. First, the user selects from a drop-down menu on the client interface, including options such as normal gait, internal rotation gait, and external rotation gait, and the user's selection is recorded. This result is the gait pattern identifier. Second, the acquired 3D foot point cloud data is processed using the Poisson reconstruction algorithm to generate a smooth closed triangular mesh surface model. Then, the curvature of each part of the mesh surface model is calculated, and the curvature distribution information representing key areas such as the arch dome and forefoot transverse arch is extracted. This distribution information constitutes the foot surface contour feature information. Third, the user wears smart insoles with integrated inertial measurement units (IMUs) or has reflective markers affixed to their feet and walks. The corresponding receiving device records the continuous 3D angle and angular velocity data of the foot during walking. This dataset arranged in chronological order is the foot motion sequence information.

[0094] Next, the gait pattern identifier classification label is converted into a sparse binary vector using one-hot encoding. Then, through vector concatenation, this binary vector is simply joined end-to-end with the generated initial foot biomechanical feature expression to form a new, longer combined vector, which is the intermediate biomechanical feature expression. Afterward, principal component analysis is used to process the acquired foot motion sequence information. Principal component analysis is performed on the angle change sequence of the foot in three-dimensional space to find the first two or three principal component directions that best explain the motion variation. These directions usually correspond to the main movements of the foot, such as inversion / eversion, dorsiflexion, and plantarflexion. Then, the projection range of the original motion data in each principal component direction is calculated, and the absolute value of the difference between the maximum and minimum values ​​of the range is taken as the dominant motion amplitude in that direction.

[0095] Subsequently, through vector addition, the calculated amplitudes of multiple dominant movements are added to the corresponding values ​​at the intermediate mechanical feature expressions, injecting dynamic movement range information into the existing features. This results in an enhanced mechanical feature expression with richer information dimensions and dynamic characteristics. Finally, a feature fusion network processes the acquired foot surface contour feature information, such as a set of curvature values ​​and the obtained enhanced mechanical feature expressions. For example, a cross-attention network uses the former as the query and the latter as the key and value, calculates the association weight between the two, and deeply fuses the information of the enhanced mechanical feature expressions with the contour feature information according to the weights. The final output is a new, unified feature vector that integrates all information of static dimensions, gait patterns, dynamic amplitudes, and surface morphology. This final feature vector is the foot biomechanical data used in subsequent steps.

[0096] For example, following the specific implementation of the previous step, after obtaining the user's foot size data, Company A's platform begins to generate foot biomechanical data. First, it inputs the user's foot length, foot width, and other dimensional components into a neural network model in its backend to quickly calculate an initial foot biomechanical feature expression containing information such as the estimated arch load-bearing capacity. At the same time, the platform displays a gait pattern questionnaire to the user, who selects the mild pronation option based on self-awareness, generating a gait pattern identifier. In addition, the platform performs surface reconstruction on the user's foot scan point cloud, automatically calculates that the arch curvature radius is moderate, and extracts it as foot surface contour feature information.

[0097] Next, the user walks several steps on a test mat equipped with IMU sensors, either wearing test socks containing IMU sensors or barefoot. The data recorded by the IMU sensors is uploaded as foot movement sequence information. Subsequently, the platform encodes the slight pronation and splices it with the initial biomechanical features to form an intermediate biomechanical feature representation. Then, the IMU data is analyzed to calculate the main range of motion angles of the user's foot in the pronation and supination directions, obtain the dominant motion amplitude, and superimpose it into the intermediate features to form an enhanced biomechanical feature representation. Finally, a fusion network deeply combines this enhanced representation with the user's arch curvature contour features to output a unique and comprehensive foot biomechanical data file.

[0098] This step deeply integrates the user's biomechanical characteristics and behavioral habits, providing a core basis for the subsequent generation of a foot surface that combines precise functional adaptation and personalized support.

[0099] Step 103: Extract contour features and curvature features from the style reference image, and convert the contour features and curvature features into geometric change information.

[0100] Optionally, step 103 may specifically include:

[0101] Step 1031: Identify the target shape boundary lines in the style reference image to obtain contour feature information. The target shape boundary lines include the outer contour lines and the main internal segmentation lines of the object depicted in the style reference image.

[0102] Step 1032: Measure the curvature change at each point on the boundary line of the target shape to obtain the curvature feature information of the style reference image.

[0103] Step 1033: Establish the feature correspondence between the style reference image and the preset standard three-dimensional foot basic shape.

[0104] Step 1034: Based on the feature correspondence, the morphological trend of the target shape boundary line described by the contour feature information and the curvature change described by the curvature feature information are jointly mapped into geometric change information.

[0105] In this step, the outline features refer to the set of lines extracted from the style reference image that describe its core shape skeleton, used to provide a macroscopic morphological style tone for the sole surface.

[0106] Curvature features refer to data that quantifies the degree of local curvature variation of the aforementioned contour lines, used to provide microscopic details of the arc and transition style of the sole surface.

[0107] Geometric transformation information refers to a set of digital instructions that can guide the basic shape of a three-dimensional foot to undergo specific deformations to conform to stylistic features. It is used to convert stylistic elements of a two-dimensional image into control parameters that can be applied to a three-dimensional model.

[0108] The target shape boundary lines refer to the coherent lines identified in the style reference image that define the main visual form of the object being depicted. These include the outer contour lines that outline the overall range of the object and the main internal segmentation lines that divide the core structure of the object. They serve as a precise carrier for feature extraction and are obtained through a combination of advanced edge detection and semantic segmentation.

[0109] Curvature variation refers to the degree of drastic change in direction at each point on the boundary line of the target shape, i.e., curvature value, used to quantify the roundness or sharpness of the style, and is obtained by calculating the first and second derivatives of the line.

[0110] The preset standard three-dimensional foot base shape refers to a neutral, unstylized, universal three-dimensional foot surface model, which serves as a unified geometric carrier for style transfer.

[0111] Feature correspondence refers to the mapping link established between two-dimensional feature points in the style reference image and the three-dimensional surface area of ​​the standard three-dimensional foot base shape. It is used to establish a bridge for cross-dimensional editing and is obtained through correlation analysis between image features and the UV parameter space of the three-dimensional model.

[0112] Morphological trends refer to the overall shape direction, proportional relationships, and compositional intention implied by contour feature information. They are used to guide the overall direction of three-dimensional shape adjustment and are perceived through global analysis of the contour line set.

[0113] In this step, the style reference image is first extracted at the pixel level using the Canny edge detection algorithm. Then, the outermost closed curve is retrieved from the edge image using the contour search algorithm as the outer contour line. At the same time, the main component regions of the target object in the image are identified by the U-Net semantic segmentation model, and the boundaries of these regions are extracted as the main internal segmentation lines. Finally, through line vectorization processing, these lines are converted into a mathematical expression composed of an ordered sequence of coordinate points to obtain contour feature information.

[0114] Secondly, the numerical differential method is used to process each line point sequence in the contour feature information. The first and second derivatives of each point sequence are calculated using the central difference formula. Then, the curvature value at each point is calculated according to the curvature formula. After traversing all line point sequences, a curvature value sequence corresponding to each contour point is obtained. This sequence is the curvature feature information. Next, a correspondence is established through a pre-trained feature matching network. This feature matching network uses key points in the contour feature information, such as endpoints and inflection points, and their corresponding curvature feature information as style-side descriptors. It uses the predefined key points of the standard three-dimensional foot basic shape in its two-dimensional UV parameter space as geometric-side descriptors. It starts to calculate the similarity between the two descriptors and matches the most similar foot UV space key points for the style-side key points, thereby establishing the feature correspondence between the two.

[0115] Finally, a differentiable rendering and optimization framework is used to perform mapping. The standard three-dimensional foot shape is used as a deformable template. The feature correspondence is used as a constraint. The morphological trend contained in the contour feature information and the degree of curvature change quantified by the curvature feature information are set as optimization targets. The gradient descent method is used to iteratively adjust the three-dimensional coordinates of each vertex on the surface of the foot shape so that the projected contour and curvature distribution of the deformed shape under a specific viewpoint approximate the style features. After optimization, the required position adjustment amount of each vertex is recorded. This three-dimensional displacement vector field is the geometric change information.

[0116] For example, following the specific implementation of the previous step, Company A's platform first processes the sports car style reference image selected by the user; then, it uses its image processing engine to run edge detection and segmentation algorithms to accurately identify the smooth roofline and sharp grille outline of the sports car as target shape boundary lines, and vectorizes and stores them as contour feature information; next, it analyzes the curvature changes at each point on these lines, calculates the numerical sequence of the roofline curvature being gentle while the door fold line curvature is abrupt, and obtains curvature feature information; then, it calls a preset mapping model to establish a correlation between the features of the roofline in the image and the arch and back area of ​​the standard three-dimensional foot model, forming a feature correspondence; finally, based on this relationship, the smooth extension trend of the sports car's curve is transformed into an instruction to extend and lengthen the arch and back surface of the foot, and the sharp corner trend of the grille is transformed into an instruction to form an edge on the forefoot edge surface. All these instructions are combined into a complete set of geometric change information data packets.

[0117] This step uses image processing and geometric mapping techniques to quantify the shape trend of the contour and the curvature of the lines into specific geometric change information, thus providing a key shape control basis for the subsequent generation of a foot surface that combines aesthetic style and functional adaptation.

[0118] Step 104: Based on the foot biomechanics data and the geometric change information, iteratively update the randomly initialized three-dimensional data to obtain the initial three-dimensional plantar surface.

[0119] Optionally, step 104 may specifically include:

[0120] Step 1041: Combine the foot biomechanics data with the geometric change information to form a combination of conditions for guiding surface generation.

[0121] Step 1042: Based on the combination of conditions and the randomly initialized 3D data, calculate the update direction of the 3D data in the current iteration round.

[0122] Step 1043: Based on the combination of conditions, calculate the adjustment range of the 3D data along the update direction in the current iteration.

[0123] Optionally, step 1043 may specifically include the following steps: determining a constraint boundary for limiting the shape change of the three-dimensional surface based on the foot biomechanical data contained in the condition combination; identifying a target surface region in the three-dimensional data that interacts with the constraint boundary in the current iteration; calculating the expected intensity of morphological adjustment of the target surface region based on the geometric change information contained in the condition combination; generating a dynamic adjustment coefficient related to the iteration based on the sequence number of the current iteration; and calculating the adjustment magnitude of the three-dimensional data along the update direction in the current iteration based on the relative relationship between the target surface region and the constraint boundary, the expected intensity of morphological adjustment, and the dynamic adjustment coefficient.

[0124] Step 1044: Update the three-dimensional data according to the update direction and the adjustment range to obtain the updated three-dimensional data.

[0125] Step 1045: Determine whether the updated 3D data meets the preset convergence condition. If the updated 3D data meets the convergence condition, then the updated 3D data is output as the initial 3D foot surface.

[0126] Step 1046: If the updated 3D data does not meet the convergence condition, the updated 3D data is used as the 3D data in the next iteration round. The steps of calculating the update direction, calculating the adjustment range, updating the 3D data, and determining whether the convergence condition is met are repeated until the convergence condition is met to output the initial 3D foot surface.

[0127] In this step, the initial 3D foot surface refers to the 3D surface model that is finally generated through the iterative update process, which incorporates foot biomechanical constraints and visual style elements, and is used as the benchmark model for subsequent fine-tuning.

[0128] Conditional combination refers to a unified control signal formed by integrating foot biomechanical data and geometric change information. It is used to guide the surface to meet both functional and aesthetic requirements during the generation process and is obtained through vector splicing or conditional coding techniques.

[0129] Constraint boundaries refer to the virtual boundaries implicit in foot biomechanical data that define the shape of the foot surface in three-dimensional space and cannot be crossed. They are used to ensure the biomechanical rationality of the generated surface and are calculated by analyzing the pressure distribution and stability requirements in the biomechanical data.

[0130] The target surface region refers to the local surface parts in the 3D data of the current iteration that come into contact with, penetrate, or are too close to the constraint boundary. These are key areas that need to be shaped to meet the constraints and are identified by the spatial distance detection algorithm.

[0131] The expected intensity of shape adjustment refers to the quantified value of the degree of shape change expected to be achieved for the target surface region based on geometric change information. It is used to control the intensity of style transfer and is obtained by calculating the difference between the current surface shape and the style target shape.

[0132] The dynamic adjustment coefficient is a numerical factor that decays or adjusts as the iteration round number changes. It is used to allow for larger adjustments in the early stages of the iteration and to perform fine optimizations in the later stages of the iteration to achieve stable convergence. It is calculated based on the current round using a preset decay function.

[0133] Updated 3D data refers to the new 3D point set or surface representation obtained by modifying the original 3D data according to the update direction and adjustment range in the current iteration round, which is used as the input or final output of the next iteration round.

[0134] The preset convergence criteria refer to a set of standards used to determine whether the iterative process can be terminated, and are used to control the timing of the end of the generation process.

[0135] In this step, foot biomechanical data and geometric change information are first combined through a conditional encoder network. The conditional encoder network receives these two inputs and fuses and encodes them through a fully connected layer and a nonlinear activation function, outputting a unified, fixed-dimensional conditional vector. This conditional vector is the conditional combination used to guide the entire generation process.

[0136] Secondly, the update direction is calculated through a predictor network of a denoising diffusion model. The current iteration's 3D data is used as noise data and the generated condition combination is used as input. The difference between the current 3D data and the target state described by the condition combination is analyzed, and a vector field with the same dimension as the current 3D data is predicted. This vector field indicates the direction in which each point in the current 3D data should move to get closer to the target state. This direction vector is the update direction.

[0137] Next, the adjustment range is determined through a series of calculations. First, by analyzing the information on the maximum pressure threshold and the safe range of joint angles in the foot biomechanical data, a series of insurmountable curved surface boundaries, i.e., constraint boundaries, are defined in three-dimensional space. Then, by calculating the signed distance field between the surface represented by the current three-dimensional data and these constraint boundaries, regions with negative distance values ​​or less than the safe threshold are identified. These regions are the target curved surface regions. Subsequently, by calculating the curvature difference or normal difference between the target curved surface region under the current geometric shape and the ideal style shape described by the geometric change information, the difference value is quantified as the expected intensity of shape adjustment. At the same time, a preset exponential decay function is used to calculate a dynamic adjustment coefficient that changes from large to small according to the sequence number of the current iteration round. Finally, a comprehensive calculation function is used to calculate a unified adjustment range scalar value or an amplitude vector with the same dimension as the update direction for the entire current three-dimensional data, taking the relative distance relationship between the target curved surface region and the constraint boundary, the expected intensity of shape adjustment, and the dynamic adjustment coefficient as input.

[0138] Then, the 3D data is updated through vector scaling and addition operations. The calculated update direction vector is multiplied element-wise with the calculated adjustment magnitude value to obtain the actual position adjustment amount. This adjustment amount is then added to the coordinates of the corresponding point in the current 3D data to obtain the updated 3D data. Afterward, an evaluation algorithm is used to determine whether convergence has been achieved. The root mean square value of the position change of the updated 3D data and the 3D data before the update at all corresponding points is calculated. This root mean square value is compared with a preset, very small positive threshold, or it is checked whether the current iteration has reached the preset maximum iteration. If the change amount is less than the threshold or the maximum iteration has been reached, the preset convergence condition is determined to be met. The updated 3D data is then converted into a triangular mesh or NURBS surface format and output as the initial 3D foot surface.

[0139] Finally, the loop control logic is used for iteration. If the convergence condition is not met, the updated 3D data is used as the new current 3D data, and the iteration process is repeated to start the next round of iteration. This loop continues until the convergence condition is met. At this time, the loop terminates and the final initial 3D foot surface is output.

[0140] For example, following the specific implementation of the previous step, the generation engine of Company A's platform is started. First, the user's foot biomechanics data and the geometric change information of the sports car style are input into a condition fusion module to generate a unified condition combination. Then, the iteration begins: starting with an initial 1 million random point clouds, a prediction network is used to analyze the difference between the current point cloud and the condition combination, and calculate which direction each point should move to update the direction. At the same time, according to the biomechanical data, a constraint boundary is defined, such as the lowest point of the arch not being lower than a certain height, and the area in the current point cloud that violates the constraint boundary is identified as the target curved surface area.

[0141] Next, based on the style instructions, the desired smoothness of these target surface regions is calculated. Considering the current iteration number (e.g., the 10th iteration), the step size for this movement is calculated to be relatively large. Then, the direction and step size are combined to move all points, resulting in a new point cloud. It is then determined whether the shape of the new point cloud has stabilized and is no longer changing drastically. Since it is an early iteration, the changes are still large, so the convergence condition is not met. Thus, the new point cloud is used as input to start the next iteration. This process is repeated until, after hundreds of iterations, the shape of the point cloud changes to negligible levels, and convergence is determined. This point cloud, which has now formed a clear, smooth shape and combines the biomechanical support characteristics of the user's foot with the streamlined shape of a sports car, is converted into a three-dimensional mesh model, i.e., the initial three-dimensional foot surface.

[0142] This step achieves the coordinated generation of functionality and artistry, ensuring that the final surface is not a simple superposition of form and function, but rather an optimal geometric shape that can simultaneously meet both requirements, found through repeated iterations. This provides a high-quality and rational initial design for subsequent personalized fine-tuning.

[0143] Step 105: In the intermediate expression space corresponding to the initial three-dimensional plantar surface, the first type of intermediate variables that determine the appearance and the second type of intermediate variables that determine the mechanical properties are distinguished and controlled. By modifying the first type of intermediate variables, the target parameterized plantar surface is obtained.

[0144] Optionally, step 105 may specifically include:

[0145] Step 1051: Perform intermediate form transformation on the initial three-dimensional foot sole surface to obtain the spatial state representation of the initial three-dimensional foot sole surface in the intermediate expression space.

[0146] Step 1052: From the spatial state representation, separate the first subset corresponding to the first type of intermediate variables and separate the second subset corresponding to the second type of intermediate variables.

[0147] Optionally, step 1052 may specifically include the following steps: constructing an association network for the spatial state representation, wherein the nodes of the association network correspond to each intermediate variable in the spatial state representation, and the edges of the association network represent the interaction strength between the corresponding intermediate variables; in the association network, according to the foot surface attributes described by the data content corresponding to the intermediate variables, labeling each intermediate variable with a category label indicating whether the intermediate variable belongs to a first attribute or a second attribute, to obtain first attribute intermediate variables and second attribute intermediate variables; in the association network, identifying a first sub-network region formed by interconnecting first attribute intermediate variables and a second sub-network region formed by interconnecting second attribute intermediate variables; extracting the set of intermediate variables corresponding to the nodes of the first sub-network region from the spatial state representation to form a first subset; extracting the set of intermediate variables corresponding to the nodes of the second sub-network region from the spatial state representation to form a second subset.

[0148] Step 1053: Receive editing input information for the first type of intermediate variable.

[0149] Step 1054: Based on the edit input information, adjust the content of the intermediate variables contained in the first subset to form the adjusted first subset.

[0150] Step 1055: Combine the adjusted first subset with the second subset to form an updated spatial state representation.

[0151] Step 1056: Perform a reverse transformation on the updated spatial state representation to generate the target parameterized plantar surface.

[0152] In this step, intermediate representation space refers to a mathematical space or data structure that is more abstract and easier to separate than the original 3D model representation, used to decouple complex surfaces.

[0153] The first type of intermediate variables refers to the set of parameters in the intermediate expression space that mainly affect the visual appearance of the surface. These parameters are used to independently control the aesthetic style of the surface and are obtained by filtering from the spatial state representation through a separation algorithm.

[0154] The second type of intermediate variables refers to the set of parameters in the intermediate expression space that mainly determine the biomechanical support properties of the surface, and are used to maintain the functional stability of the surface.

[0155] The target parameterized foot surface refers to the final three-dimensional surface model obtained after adjusting the first type of intermediate variables, which updates the appearance while maintaining the original mechanical properties, and is used as the result of the design completion.

[0156] Spatial state representation refers to the specific mathematical description of the initial three-dimensional foot surface in the intermediate expression space. It is usually a high-dimensional vector or feature tensor used to carry all the encoded information of the surface, and is obtained through the transformation by the encoder network.

[0157] The first subset refers to the data set separated from the spatial state representation that contains only the first type of intermediate variables. It is used to receive and respond to appearance editing operations and is extracted through a graph partitioning algorithm.

[0158] The second subset refers to the data set separated from the spatial state representation that contains only the second type of intermediate variables, and is used to maintain the mechanical properties by keeping them unchanged during the editing process.

[0159] A relational network is a graph structure that describes the relationships between intermediate variables in a spatial state representation. It is used to analyze and separate variables with different attributes. It is constructed by calculating the correlation matrix between variables and thresholding it.

[0160] Interaction strength refers to the weight of the edge connecting two nodes in a network of relationships. It is used to quantify the degree of association between two intermediate variables and is obtained through statistical analysis.

[0161] Foot surface properties refer to the physical or geometric characteristics related to the surface represented by intermediate variables, such as forefoot curvature, arch height, and local hardness. They are used to classify intermediate variables and are obtained by back-analyzing the role of variables in the generation process.

[0162] Category labels refer to the classification identifiers assigned to each intermediate variable, used to indicate which category it belongs to in terms of appearance or mechanical properties. They are obtained by labeling using a supervised classifier or attribute-based rules.

[0163] The first and second attribute intermediate variables refer to intermediate variables that have been labeled with the corresponding categories. They correspond to the first type of intermediate variable and the second type of intermediate variable, respectively, and are obtained directly through the labeling process.

[0164] The first and second sub-network regions refer to the subgraphs formed in the relational network by the first and second attribute intermediate variables as nodes, which are tightly connected by high-strength edges. They are used to locate clusters of appearance control variables and mechanical control variables, and are identified by the community detection algorithm.

[0165] The intermediate variable set refers to the entirety of intermediate variables associated with all nodes in a specific sub-network region, constituting the specific content of the first or second subset, and is extracted by index from the spatial state representation.

[0166] Editing input information refers to the command data issued by the user through the interactive interface to modify the appearance and shape, which drives the content changes of the first subset.

[0167] The adjusted first subset refers to the first subset whose internal intermediate variable values ​​have changed accordingly under the influence of the editing input information. It is used to carry the new appearance and form intention and is obtained by applying a mathematical transformation derived from the editing input information to the first subset.

[0168] The updated spatial state representation refers to the new complete intermediate representation formed by merging the adjusted first subset with the unchanged second subset. It is used to encode the new target surface and is obtained through data splicing operations.

[0169] In this step, the initial three-dimensional plantar surface is first processed by the encoder part of a pre-trained autoencoder network. The autoencoder network receives the mesh or voxel representation of the initial three-dimensional plantar surface, and gradually extracts its deep features through multi-layer convolution and fully connected operations. Finally, it outputs a fixed-length, dense feature vector, which is the spatial state representation of the surface in the intermediate representation space.

[0170] Secondly, variable separation is achieved through graph neural networks and community detection algorithms. The first step involves constructing a correlation matrix by calculating the Pearson correlation coefficients between all pairs of intermediate variables in the spatial state representation. This correlation matrix is ​​then treated as an adjacency matrix, thus constructing a network of relationships with each intermediate variable as a node and the correlation coefficient as the edge weight. The second step uses a trained classifier to analyze the role of each intermediate variable in generating the initial surface. Based on the variable's contribution to the final surface appearance and mechanical properties, the classifier assigns a category label (appearance or mechanical property) to each intermediate variable, thereby distinguishing between first-attribute and second-attribute intermediate variables. The third step utilizes the Louvain community detection algorithm... The algorithm divides the network of relationships into several communities with strong internal connections and sparse external connections based on the strength of connections between nodes. Since variables with the same attribute marked in the first step are often more strongly associated with each other, the algorithm naturally identifies the first sub-network region mainly formed by intermediate variables of the first attribute and the second sub-network region mainly formed by intermediate variables of the second attribute. In the fourth step, through simple data indexing operations, the algorithm extracts the values ​​of all intermediate variables whose node indices belong to the first sub-network region from the original spatial state representation, forming the first subset; similarly, it extracts the values ​​of all intermediate variables whose node indices belong to the second sub-network region, forming the second subset.

[0171] Next, editing input information is received through a graphical user interface. Users can adjust the style sharpness using sliders on the interface and change local contours by dragging control points directly on the curved surface. These interactive operations are captured in real time and converted into numerical adjustment instructions for the first type of intermediate variables. The set of these numerical adjustment instructions is the editing input information. Then, the first subset is adjusted according to the editing input information. The editing input information is parsed into a set of target changes, and these target changes are applied to the corresponding intermediate variables of the first subset through a predefined mapping function. For example, if the editing is to increase the overall sharpness, the mapping function will proportionally increase the values ​​of the variables that control the edge sharpness of the curved surface in the first subset, thus obtaining the adjusted first subset.

[0172] Then, the data is combined through array concatenation operations. The adjusted first subset and the unchanged second subset are reassembled according to their original index order in the spatial state representation to form a brand new and complete feature vector, which is the updated spatial state representation. Finally, the decoder part of the autoencoder network performs a reverse transformation, and the updated spatial state representation is input into the corresponding decoder network. The decoder network transforms and reconstructs this intermediate representation space vector into a specific and complete three-dimensional mesh model through a series of deconvolution or fully connected layer operations. This newly generated three-dimensional model is the target parameterized plantar surface.

[0173] For example, following the specific implementation of the previous step, after generating an initial three-dimensional foot surface on Company A's platform, the goal is to make the surface appear more dynamic without affecting its support function. Secondly, an encoder is used to compress the initial surface into a spatial state representation containing 1024 feature values. Then, the relationships between these feature values ​​are analyzed to construct a relationship network, and through analysis, 300 features that mainly affect appearance and 724 features that mainly affect geology are automatically identified.

[0174] Then, using a community discovery algorithm, the 1024 features were successfully clustered into two main groups. Subsequently, the platform interface provided a streamline enhancement slider, which, when dragged to the right, received the edit input information and converted it into a specific mathematical transformation of the first subset of 300 appearance features, changing their values ​​to form the adjusted first subset. This modified set of appearance features was then merged with the second subset of 724 mechanical features that remained unchanged, resulting in a new updated spatial state representation containing 1024 feature values. Finally, the decoder was called to decode this new representation, instantly generating a new surface. This new surface retains the original comfort support characteristics, making the arch lines and lateral contours of the foot smoother and sharper, full of speed. This is the final target parametric plantar surface.

[0175] This step allows for free and intuitive adjustment and re-creation of the aesthetic form without compromising the core biomechanical properties of the surface. It completely solves the design rigidity problem caused by the deep coupling of function and form, greatly improves the flexibility of customized design and the freedom to realize design intentions, and ultimately efficiently outputs the target surface that fully meets personalized needs.

[0176] Figure 2 This application provides a schematic diagram of a system for automatically generating parametric plantar surfaces, as shown below. Figure 2 As shown, the system includes:

[0177] The acquisition module 21 is used to acquire the target user's foot size data and style reference image, as well as a set of randomly initialized 3D data;

[0178] Generation module 22 is used to generate foot biomechanical data based on the foot size data;

[0179] The conversion module 23 is used to extract contour features and curvature features from the style reference image and convert the contour features and curvature features into geometric transformation information;

[0180] Update module 24 is used to iteratively update the randomly initialized three-dimensional data based on the foot biomechanical data and the geometric change information to obtain an initial three-dimensional plantar surface;

[0181] Modification module 25 is used to distinguish and control the first type of intermediate variables that determine the appearance and the second type of intermediate variables that determine the mechanical properties in the intermediate expression space corresponding to the initial three-dimensional plantar surface. By modifying the first type of intermediate variables, the target parameterized plantar surface is obtained.

[0182] Figure 2 The aforementioned automatic generation system for parametric plantar surfaces can perform... Figure 1 The implementation principle and technical effects of the automatic generation method for parametric plantar surfaces described in the illustrated embodiment will not be repeated here. The specific methods by which each module and unit of the automatic generation system for parametric plantar surfaces in the above embodiments perform operations have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0183] In one possible design, Figure 2 The illustrated embodiment of an automatic generation system for parametric plantar surfaces can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0184] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0185] The processing component 32 is used for the above Figure 1 The embodiment describes an automatic generation method for parametric plantar surfaces.

[0186] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0187] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0188] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0189] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0190] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0191] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0192] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is a method for automatically generating parametric plantar surfaces.

[0193] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0194] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0195] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0196] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for automatically generating parametric foot sole surfaces, characterized in that, include: Acquire the target user's foot size data and style reference images, as well as a set of randomly initialized 3D data; Based on the foot size data, foot biomechanical data is generated; Contour features and curvature features are extracted from the style reference image, and the contour features and curvature features are converted into geometric transformation information; Based on the foot biomechanics data and the geometric change information, the randomly initialized three-dimensional data is iteratively updated to obtain the initial three-dimensional plantar surface; In the intermediate expression space corresponding to the initial three-dimensional plantar surface, the first type of intermediate variables that determine the appearance and the second type of intermediate variables that determine the mechanical properties are distinguished and controlled. By modifying the first type of intermediate variables, the target parameterized plantar surface is obtained.

2. The method according to claim 1, characterized in that, Based on the foot size data, foot biomechanical data is generated, including: Based on the multiple dimensional components contained in the foot size data, an initial foot biomechanical feature representation is generated; The gait pattern identifier of the target user, the foot surface contour feature information of the target user, and the foot movement sequence information of the target user during the movement process are obtained. The gait pattern identifier is combined with the initial foot biomechanical feature representation to form an intermediate biomechanical feature representation; Based on the foot movement sequence information, the dominant movement amplitude of the target user's foot in multiple movement directions is calculated; The dominant motion amplitude is superimposed on the intermediate mechanical feature expression to form an enhanced mechanical feature expression; The foot surface contour feature information is combined with the enhanced mechanical feature expression to generate foot biomechanical data.

3. The method according to claim 1, characterized in that, Contour features and curvature features are extracted from the style reference image, and the contour features and curvature features are converted into geometric transformation information, including: Identify the target shape boundary lines in the style reference image to obtain contour feature information. The target shape boundary lines include the outer contour lines and the main internal segmentation lines of the object depicted in the style reference image. The curvature feature information of the style reference image is obtained by measuring the curvature change at each point on the boundary line of the target shape. Establish the feature correspondence between the style reference image and the preset standard three-dimensional foot basic shape; Based on the feature correspondence, the morphological trend of the target shape boundary lines described by the contour feature information and the curvature change described by the curvature feature information are jointly mapped into geometric change information.

4. The method according to claim 1, characterized in that, Based on the foot biomechanical data and the geometric change information, the randomly initialized three-dimensional data is iteratively updated to obtain an initial three-dimensional plantar surface, including: The foot biomechanics data and the geometric change information are combined to form a combination of conditions to guide the generation of the curved surface; Based on the combination of conditions and randomly initialized 3D data, calculate the update direction of the 3D data in the current iteration round; Based on the combination of conditions, calculate the adjustment range of the 3D data along the update direction in the current iteration round; The three-dimensional data is updated according to the update direction and the adjustment range to obtain the updated three-dimensional data; Determine whether the updated 3D data meets the preset convergence condition. If the updated 3D data meets the convergence condition, then the updated 3D data is output as the initial 3D foot surface. If the updated 3D data does not meet the convergence condition, the updated 3D data is used as the 3D data in the next iteration round. The steps of calculating the update direction, calculating the adjustment range, updating the 3D data, and determining whether the convergence condition is met are repeated until the convergence condition is met to output the initial 3D foot surface.

5. The method according to claim 1, characterized in that, In the intermediate representation space corresponding to the initial three-dimensional plantar surface, the first type of intermediate variables that determine the appearance and the second type of intermediate variables that determine the mechanical properties are distinguished and controlled. By modifying the first type of intermediate variables, the target parameterized plantar surface is obtained, including: The initial three-dimensional plantar surface is transformed into an intermediate form to obtain the spatial state representation of the initial three-dimensional plantar surface in the intermediate expression space. From the spatial state representation, a first subset corresponding to the first type of intermediate variables and a second subset corresponding to the second type of intermediate variables are separated. Receive editing input information for the first type of intermediate variables; Based on the edit input information, the intermediate variables contained in the first subset are adjusted to form the adjusted first subset; The adjusted first subset is combined with the second subset to form an updated spatial state representation; The updated spatial state representation is reverse-transformed to generate the target parameterized plantar surface.

6. The method according to claim 4, characterized in that, Based on the aforementioned combination of conditions, the adjustment magnitude of the 3D data along the update direction in the current iteration is calculated, including: Based on the foot biomechanics data contained in the combination of conditions, determine the constraint boundaries used to limit the shape changes of the three-dimensional curved surface; In the current iteration, identify the target surface region in the 3D data that interacts with the constraint boundary; Based on the geometric change information contained in the combination of conditions, calculate the expected intensity of morphological adjustment for the target surface region; Based on the current iteration round number, a dynamic adjustment coefficient related to the iteration round is generated; Based on the relative relationship between the target surface region and the constraint boundary, the expected intensity of the morphological adjustment, and the dynamic adjustment coefficient, the adjustment magnitude of the 3D data along the update direction in the current iteration is calculated.

7. The method according to claim 5, characterized in that, From the spatial state representation, separating a first subset corresponding to the first type of intermediate variables and a second subset corresponding to the second type of intermediate variables includes: A relational network is constructed for the spatial state representation, wherein the nodes of the relational network correspond to the intermediate variables in the spatial state representation, and the edges of the relational network represent the interaction strength between the corresponding intermediate variables. In the network of relationships, based on the plantar surface attributes described by the data content corresponding to the intermediate variables, each intermediate variable is labeled with a category label indicating whether it belongs to the first attribute or the second attribute, thus obtaining the first attribute intermediate variable and the second attribute intermediate variable. In the network of relationships, a first sub-network region formed by interconnecting intermediate variables of the first attribute and a second sub-network region formed by interconnecting intermediate variables of the second attribute are identified. From the spatial state representation, extract the set of intermediate variables corresponding to the nodes of the first sub-network region to form the first subset; From the spatial state representation, the set of intermediate variables corresponding to the nodes of the second sub-network region is extracted to form the second subset.

8. A system for automatically generating parametric plantar surfaces, characterized in that, include: The acquisition module is used to acquire the target user's foot size data and style reference image, as well as a set of randomly initialized 3D data; A generation module is used to generate foot biomechanical data based on the foot size data; The conversion module is used to extract contour features and curvature features from the style reference image, and convert the contour features and curvature features into geometric transformation information; An update module is used to iteratively update the randomly initialized three-dimensional data based on the foot biomechanical data and the geometric change information to obtain an initial three-dimensional plantar surface; The modification module is used to distinguish and control the first type of intermediate variables that determine the appearance and the second type of intermediate variables that determine the mechanical properties in the intermediate expression space corresponding to the initial three-dimensional plantar surface. By modifying the first type of intermediate variables, the target parameterized plantar surface is obtained.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the automatic generation method of parametric plantar surface as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements a method for automatically generating parametric plantar surfaces as described in any one of claims 1 to 7.

Images