Local skeleton projection fusion method for digital garment re-fitting

Abstract

The invention discloses a local skeleton projection fusion method for digital clothing re-adaptation, which comprises the following steps of: taking a source digital role model which wears source clothing and is bound with a source skeleton system and a target digital role model which is bound with a target skeleton system as input, and generating target clothing which is matched with the posture and the body type of the target digital role model; the method comprises the following steps: constructing a local skeleton projection fusion representation, and geometrically representing a garment as linear fusion of projection results of the garment in each skeleton local coordinate system; on the basis, a local residual variable is introduced into a skeleton local coordinate space and iterative optimization is carried out, so that fine adjustment of clothing geometry is realized; meanwhile, a target function is constructed by combining contact constraint, shape keeping constraint and regularization terms, the collision relation and the fitting relation between the clothes and the target role model are optimized, clothes re-adaptation between different digital role models is achieved, and geometric detail features of the clothes are kept.

Description

A Local Skeletal Projection Fusion Method for Digital Clothing Refitting Technical Field

[0001] This invention belongs to the field of computer graphics and 3D digital content generation technology, and in particular, it is a local skeletal projection fusion method for digital clothing refitting. Background Technology

[0002] Digital clothing refitting refers to the technical process of converting a clothing model designed for a source digital character and applying it to a target digital character model with different body types, proportions, or topological structures. It is widely used in virtual try-on, game development, and digital film and television content generation. With the increasing prevalence of high-resolution clothing models and multi-layered clothing structures, accurately transferring clothing geometry between different characters has become a crucial issue in 3D modeling and animation production.

[0003] Existing methods for clothing refitting mainly include physical simulation-based methods, data-driven methods, and geometry optimization-based methods. Physical simulation-based methods typically calculate clothing deformation by constructing a mechanical model, but this involves a large amount of computation and is highly dependent on human body parameter models. Data-driven methods rely on training data for deformation prediction, and their generalization ability is limited by the distribution of training samples. Geometry optimization-based methods typically perform fitting calculations in the global vertex coordinate space, which offers a certain degree of controllability, but when dealing with large-scale differences in body shape or posture, they are prone to optimization instability or loss of detail. Recent research, such as the paper "Zizhou Huang, Chrystiano Araújo, Andrew Kunz, Denis Zorin, Daniele Panozzo, and Victor Zordan. 2025. Intersection-free Garment Retargeting. In Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Papers. 1–12," reduces the corresponding requirements through skeleton embedding technology. However, the rigid skeleton assignment it relies on is prone to generating semantically incorrect connections under non-standard poses, making subsequent optimization extremely sensitive to initial parameters.

[0004] Therefore, improving the stability of the optimization process while ensuring computational efficiency, and maintaining the local geometric features and fit of clothing between different roles, remains a critical technical problem that needs to be solved in the field of clothing refitting. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing clothing refitting technologies, such as poor numerical stability, limited generalization ability, and inability to effectively maintain original design details when dealing with significant differences in body shape or posture. This invention provides a local skeletal projection fusion method for digital clothing refitting.

[0006] The technical solution for achieving the objective of this invention is: a local skeletal projection fusion method for digital clothing refitting, the method comprising the following steps:

[0007] Step 1: Obtain source data and target data; wherein, the source data includes a source digital character model and its bound source skeletal system and source clothing, and the target data includes a target digital character model and its bound target skeletal system; the source clothing has a three-dimensional geometric representation, including multiple vertices;

[0008] Step 2: Define local coordinate systems for each bone in the source and target skeletal systems, respectively;

[0009] Step 3: Based on the projection function, calculate the local projection coordinates of each vertex of the source clothing in the local coordinate system of each bone in the source skeleton system;

[0010] Step 4: Map the local projection coordinates to the semantic bone local coordinate system corresponding to the target bone system, and transform them to the global space through the inverse projection function to obtain the candidate global position set of each vertex for each bone in the target bone system;

[0011] Step 5: For each vertex, perform linear weighted fusion of the candidate global positions of its corresponding bones to determine the initial fitting position of the vertex in the target space, thereby obtaining the preliminary fitting target clothing.

[0012] Step 6: Based on the initial adaptation position, the final adaptation target clothing is generated by optimizing the coordinate residuals and fusion weight residuals in the local bone projection space.

[0013] Furthermore, in step 1, the source skeletal system and the target skeletal system have similar topological structures, specifically including:

[0014] If the topological structures are consistent, establish a one-to-one semantic relationship between the bones of the source skeletal system and the target skeletal system;

[0015] If there are differences in the topology, the skeletal structure is adjusted through preset pruning or completion operations until the one-to-one semantic relationship is established.

[0016] Furthermore, the process of constructing the local coordinate system in step 2 includes:

[0017] Step 2-1: Using the starting joint of the bone as the origin of the local coordinate system, normalize the direction vector from the starting joint to the ending joint and use it as the z-axis direction of the local coordinate system.

[0018] Step 2-2: Select a reference vector or parent bone basis vector, eliminate its z-axis component through orthogonalization and normalize it to serve as the x-axis direction of the local coordinate system;

[0019] Steps 2-3: Determine the direction of the y-axis by performing the cross product of the z-axis and x-axis, thereby constructing a local coordinate system that satisfies the right-hand orthogonal constraint.

[0020] Furthermore, step 2-2 specifically includes:

[0021] When the skeleton is the root skeleton, a reference vector that is not collinear with the z-axis direction is selected, the reference vector is orthogonalized to eliminate its component in the z-axis direction, and after normalization, it is taken as the x-axis direction.

[0022] When a bone has a parent bone, a basis vector in the local coordinate system of the parent bone is selected as the initial direction, orthogonalized and normalized to obtain the x-axis direction of the current bone.

[0023] Specifically, when the initial direction is approximately parallel to the z-axis direction of the current bone, another basis vector in the local coordinate system of the parent bone is selected as a backup direction and the orthogonalization process is re-executed.

[0024] Furthermore, step 3 specifically includes:

[0025] Step 3-1: For each vertex of the source garment, calculate its displacement vector relative to the starting joint of the skeleton.

[0026] Step 3-2: Based on the projection function, the displacement vector is projected onto the three orthogonal basis vector directions of the local coordinate system of the skeleton, and the three directional components are obtained by dot product operation;

[0027] Step 3-3: Divide the three directional components by the corresponding bone length for normalization to construct scale-independent local projection coordinates;

[0028] Steps 3-1 to 3-3 are performed on each bone in the source skeleton system to construct the projection representation of the vertex in the multi-bone local coordinate space.

[0029] Furthermore, step 4 specifically includes:

[0030] Step 4-1: Scale the local projected coordinates using the target bone length; the target bone is the bone in the target bone system corresponding to each bone in the source bone system.

[0031] Step 4-2: Use the basis vectors of the target bone to rotate and synthesize the restored coordinates to generate a displacement vector relative to the origin of the bone.

[0032] Step 4-3: Superimpose the displacement vector onto the starting joint of the target bone to obtain the candidate global position of the vertex relative to the target bone;

[0033] Steps 4-1 to 4-3 are performed on each bone in the target skeletal system to obtain a set of candidate global positions of the vertex for each bone in the target skeletal system.

[0034] Furthermore, step 5 determines the initial adaptation position, and the specific process includes:

[0035] Step 5-1: For each vertex of the source clothing, a set of skinning weights pre-associated with it is inherited as a set of initial fusion weights under the target skeleton system; where it is necessary to ensure that all associated bones of the vertex correspond to an initial fusion weight.

[0036] Step 5-2: For each vertex, multiply the candidate global positions of each bone obtained in Step 4 by the corresponding initial fusion weights and sum them to obtain the initial adaptation position of the vertex in the target space; wherein, during the fusion process, the sum of the initial fusion weights of all associated bones of each vertex must be 1.

[0037] Furthermore, step 6 specifically includes the following processes:

[0038] Step 6-1: For each vertex, add its local projected coordinates in the local coordinate system of each bone obtained in step 3 to the local coordinate residuals of the vertex relative to each bone to obtain the perturbed local coordinates in the semantic bone local coordinate system.

[0039] Step 6-2: Introduce unconstrained variables and combine them with bounded mapping functions to calculate the fusion weight residuals;

[0040] Step 6-3: For each initial fusion weight, sum it with the fusion weight residual and normalize it to obtain the updated fusion weight;

[0041] Step 6-4: Based on the perturbed local coordinates and the updated fusion weights, iterative optimization is performed using the target loss function to obtain the updated positions of each vertex of the initially adapted target clothing.

[0042] Furthermore, the target loss function includes contact constraints. Shape retention constraints and regularization terms , is represented as: In the formula, Represents the target loss function. This represents a constraint function that controls the penetration between clothing and characters and applies user-specified clothing fit effects. This represents a constraint function that maintains consistency in the shape and fold details of the garment. This represents a constraint function that performs regularization on the optimization variables. These are the coefficients of the constraint terms.

[0045] Furthermore, in the optimization process of step 6, a coarse-to-fine hierarchical optimization strategy and a dynamic update mechanism for contact relationships are adopted to optimize the target loss function. It gradually converges to the minimum value;

[0046] The coarse-to-fine hierarchical optimization strategy specifically includes:

[0047] Preliminary matching optimization is performed on the downsampled low-resolution grid;

[0048] The initial matching optimization results are upsampled to a high-resolution grid, and refined based on the target loss function to obtain high-resolution clothing that matches the target digital character model.

[0049] The dynamic update mechanism for contact relationships specifically includes:

[0050] For each vertex of the initially adapted target clothing, the nearest bone segment is selected as the guide bone;

[0051] Project rays along the orthogonal direction of the bone segments and intersect with the target digital character model to determine the initial contact point;

[0052] After the initial adaptation of the target clothing vertices converges to the surface neighborhood of the target digital character model, the nearest neighbor search method is switched to update the contact correspondence between the clothing and the target digital character model.

[0053] Compared with the prior art, the significant advantages of this invention are:

[0054] (1) This invention constructs a local skeleton projection fusion representation, which processes clothing geometry in the local coordinate space of the skeleton. Compared with direct optimization in the global vertex coordinate space, this improves the parameterization conditions and enhances the numerical stability.

[0055] (2) The present invention introduces residual variables in the local skeletal coordinate space for iterative optimization, which can make fine adjustments to the local geometry while maintaining the overall uniformity of the garment shape.

[0056] (3) By adopting scale-independent local projection coordinates (normalization processing), the present invention effectively eliminates the influence of differences in the skeletal scale of different characters, enabling efficient clothing refitting between digital characters of different body types and postures.

[0057] (4) The present invention incorporates shape preservation constraints (such as Laplace terms, bending and curvature constraints) to ensure that the geometric details, fold features and silhouette style of the source garment can be fully preserved during the refit process.

[0058] (5) This invention helps to expand the reasonable solution space and prevent the weights from fluctuating drastically during the optimization process by using a linear hybrid structure of fusion weights and a bounded constraint mechanism of weight residuals.

[0059] (6) By introducing contact constraints and dynamic update mechanisms, this invention can effectively handle the collision and interweaving between clothing and character models, and maintain a specific fit style according to the area specified by the user.

[0060] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0061] Figure 1 is a flowchart of the local skeletal projection fusion method for digital clothing refitting according to the present invention.

[0062] Figure 2 is a schematic diagram of the local bone projection fusion method of the present invention.

[0063] Figure 3 is an input diagram of source and target data in one embodiment, where (a) is the skeletal system bound to the source digital character model, (b) is the source digital character model wearing the source clothing, (c) is the skeletal system bound to the target digital character model, and (d) is the target digital character model.

[0064] Figure 4 is a schematic diagram of an initial target garment and target skeleton system in one embodiment.

[0065] Figure 5 shows the output diagram of a target digital character model wearing the appropriate target clothing in one embodiment. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0067] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0068] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0069] In one embodiment, referring to Figure 1, a local skeletal projection fusion method for digital clothing refitting is provided, which takes a source digital character model wearing the source clothing and bound to a source skeletal system and a target digital character model bound to a target skeletal system as inputs to generate a target clothing that is adapted to the posture and body shape of the target digital character model.

[0070] The method includes the following steps:

[0071] Step 1: Obtain source data and target data; wherein, the source data includes a source digital character model and its bound source skeletal system and source clothing, and the target data includes a target digital character model and its bound target skeletal system; the source clothing has a three-dimensional geometric representation, including multiple vertices;

[0072] Here, the source clothing may also be, but is not limited to, a point cloud model, a symbolic distance field (SDF) representation, or a geometric representation that can be converted into a mesh model (Mesh) through a preset algorithm.

[0073] Step 2: Define local coordinate systems for each bone in the source and target skeletal systems, respectively;

[0074] Step 3: Based on the projection function, calculate the local projection coordinates of each vertex of the source clothing in the local coordinate system of each bone in the source skeleton system;

[0075] Step 4: Map the local projection coordinates to the semantic bone local coordinate system corresponding to the target bone system, and transform them to the global space through the inverse projection function to obtain the candidate global position set of each vertex for each bone in the target bone system;

[0076] Step 5: For each vertex, perform linear weighted fusion of the candidate global positions of its corresponding bones to determine the initial fitting position of the vertex in the target space, thereby obtaining the preliminary fitting target clothing.

[0077] Step 6: Based on the initial adaptation position, the final adaptation target clothing is generated by optimizing the coordinate residuals and fusion weight residuals in the local bone projection space.

[0078] Furthermore, in one embodiment, in step 1, the source skeletal system and the target skeletal system have a similar topological structure, specifically including:

[0079] If the topological structures are consistent, establish a one-to-one semantic relationship between the bones of the source skeletal system and the target skeletal system;

[0080] If there are differences in the topology, the skeletal structure is adjusted through preset pruning or completion operations until the one-to-one semantic relationship is established.

[0081] Furthermore, in one embodiment, the process of constructing the local coordinate system in step 2 includes:

[0082] Step 2-1: Using the starting joint of the bone as the origin of the local coordinate system, normalize the direction vector from the starting joint to the ending joint and use it as the z-axis direction of the local coordinate system.

[0083] Step 2-2: Select a reference vector or parent bone basis vector, eliminate its z-axis component through orthogonalization and normalize it to serve as the x-axis direction of the local coordinate system;

[0084] Steps 2-3: Determine the direction of the y-axis by performing the cross product of the z-axis and x-axis, thereby constructing a local coordinate system that satisfies the right-hand orthogonal constraint.

[0085] Preferably, in some embodiments, step 2-2 specifically includes:

[0086] When the skeleton is the root skeleton, a reference vector that is not collinear with the z-axis direction is selected, the reference vector is orthogonalized to eliminate its component in the z-axis direction, and after normalization, it is taken as the x-axis direction.

[0087] When a bone has a parent bone, a basis vector in the local coordinate system of the parent bone is selected as the initial direction, orthogonalized and normalized to obtain the x-axis direction of the current bone.

[0088] Specifically, when the initial direction is approximately parallel to the z-axis direction of the current bone, another basis vector in the local coordinate system of the parent bone is selected as a backup direction and the orthogonalization process is re-executed.

[0089] Furthermore, in one embodiment, step 3 specifically includes:

[0090] Step 3-1: For each vertex of the source garment, calculate its displacement vector relative to the starting joint of the skeleton.

[0091] Step 3-2: Based on the projection function, the displacement vector is projected onto the three orthogonal basis vector directions of the local coordinate system of the skeleton, and the three directional components are obtained by dot product operation;

[0092] Step 3-3: Divide the three directional components by the corresponding bone length for normalization to construct scale-independent local projection coordinates;

[0093] Steps 3-1 to 3-3 are performed on each bone in the source skeleton system to construct the projection representation of the vertex in the multi-bone local coordinate space.

[0094] Here, the local projected coordinates are expressed as :

[0095]

[0096] in, express and dot product, Represents the projection function. Indicates source clothing Each vertex in the, Represents the source skeletal system middle bones The starting joint point, This represents the local coordinate system of the source skeleton obtained in step 2. Represents bones The length.

[0097] Furthermore, in one embodiment, step 4 specifically includes:

[0098] Step 4-1: Scale the local projected coordinates using the target bone length; the target bone is the bone in the target bone system corresponding to each bone in the source bone system.

[0099] Step 4-2: Use the basis vectors of the target bone to rotate and synthesize the restored coordinates to generate a displacement vector relative to the origin of the bone.

[0100] Step 4-3: Superimpose the displacement vector onto the starting joint of the target bone to obtain the candidate global position of the vertex relative to the target bone;

[0101] Steps 4-1 to 4-3 are performed on each bone in the target skeletal system to obtain a set of candidate global positions of the vertex for each bone in the target skeletal system.

[0102] Here, the candidate global position is represented as :

[0103]

[0104] In the formula, Represents the inverse projection function. Represents the target skeletal system Mid-target skeleton The starting joint point, Represents the target skeleton Length, Represents the target skeleton The basis vectors.

[0105] Furthermore, in one embodiment, step 5, determining the initial adaptation position, specifically includes the following process:

[0106] Step 5-1: For each vertex of the source garment, a set of skinning weights pre-associated with it is inherited as a set of initial fusion weights under the target skeletal system; it is necessary to ensure that each vertex has an associated bone corresponding to an initial fusion weight. ;

[0107] Step 5-2: For each vertex, multiply the candidate global positions of each bone obtained in Step 4 by the corresponding initial fusion weights and sum them to obtain the initial adaptation position of the vertex in the target space; wherein, during the fusion process, the sum of the initial fusion weights of all associated bones of each vertex must be 1.

[0108] Here, the initial adaptation position Represented as:

[0109]

[0110] The fusion weights satisfy B represents the source skeletal system. or target skeletal system 'b' represents bone. or target skeleton .

[0111] Furthermore, in one embodiment, step 6 specifically includes the following process:

[0112] Step 6-1: For each vertex, add its local projected coordinates in the local coordinate system of each bone obtained in step 3 to the local coordinate residuals of the vertex relative to each bone to obtain the perturbed local coordinates in the semantic bone local coordinate system.

[0113] Step 6-2: Introduce unconstrained variables and combine them with bounded mapping functions to calculate the fusion weight residuals;

[0114] Step 6-3: For each initial fusion weight, sum it with the fusion weight residual and normalize it to obtain the updated fusion weight;

[0115] Step 6-4: Based on the perturbed local coordinates and the updated fusion weights, iterative optimization is performed using the target loss function to obtain the updated positions of each vertex of the initially adapted target clothing.

[0116] Preferably, in some embodiments, in step 6-2, an unconstrained variable is introduced. Through bounded mapping functions Calculate the fusion weight residuals to limit the weight adjustment amount to the coefficients. Within a defined preset range.

[0117] Preferably, in some embodiments, in step 6-3, the initial fusion weights are... With the corresponding weighted residual Adding them together gives The updated fusion weights are obtained after normalization: .

[0118] Preferably, in some embodiments, the target loss function includes contact constraints. Shape retention constraints and regularization terms , is represented as: In the formula, Represents the target loss function. This represents a constraint function that controls the penetration between clothing and characters and applies user-specified clothing fit effects. This represents a constraint function that maintains consistency in the shape and fold details of the garment. This represents a constraint function that performs regularization on the optimization variables. These are the coefficients of the constraint terms.

[0121] Preferably, in some embodiments, during the optimization process in step 6, a coarse-to-fine hierarchical optimization strategy and a dynamic update mechanism for contact relationships are adopted to optimize the target loss function. It gradually converges to the minimum value;

[0122] The coarse-to-fine hierarchical optimization strategy specifically includes:

[0123] Preliminary matching optimization is performed on the downsampled low-resolution grid;

[0124] The initial matching optimization results are upsampled to a high-resolution grid, and refined based on the target loss function to obtain high-resolution clothing that matches the target digital character model.

[0125] The dynamic update mechanism for contact relationships specifically includes:

[0126] For each vertex of the initially adapted target clothing, the nearest bone segment is selected as the guide bone;

[0127] Project rays along the orthogonal direction of the bone segments and intersect with the target digital character model to determine the initial contact point;

[0128] After the initial adaptation of the target clothing vertices converges to the surface neighborhood of the target digital character model, the nearest neighbor search method is switched to update the contact correspondence between the clothing and the target digital character model.

[0129] This method constructs a local skeletal projection fusion representation, which represents the clothing geometry as a linear fusion of its projection results in the local coordinate systems of each skeleton. Based on this, it introduces local residual variables in the local coordinate space of the skeleton and performs iterative optimization to achieve fine adjustment of the clothing geometry. At the same time, it constructs an objective function by combining contact constraints, shape preservation constraints and regularization terms to optimize the collision and fitting relationships between the clothing and the target character model, thereby achieving clothing refitting between different digital character models while maintaining the detailed features of the clothing geometry.

[0130] In one embodiment, a local skeletal projection fusion system for digital clothing refitting is provided, the system comprising:

[0131] The first module is used to: acquire source data and target data; wherein, the source data includes a source digital character model and its bound source skeletal system and source clothing, and the target data includes a target digital character model and its bound target skeletal system; the source clothing has a three-dimensional geometric representation, including multiple vertices;

[0132] The second module is used to define local coordinate systems for each bone in both the source and target skeletal systems.

[0133] The third module is used to: calculate the local projection coordinates of each vertex of the source clothing in the local coordinate system of each bone in the source skeleton system based on the projection function;

[0134] The fourth module is used to: map the local projection coordinates to the semantic bone local coordinate system corresponding to the target bone system, and transform them to the global space through the inverse projection function to obtain the candidate global position set of each vertex for each bone in the target bone system;

[0135] The fifth module is used to: for each vertex, perform linear weighted fusion of the candidate global positions of each corresponding bone to determine the initial adaptation position of the vertex in the target space, thereby obtaining the initial adaptation target clothing;

[0136] The sixth module is used to generate the final adapted target clothing by optimizing the coordinate residuals and fusion weight residuals in the local bone projection space based on the initial adaptation position.

[0137] Specific limitations regarding the local skeletal projection fusion system for digital clothing refitting can be found in the limitations of the local skeletal projection fusion method for digital clothing refitting described above, and will not be repeated here. Each module in the aforementioned local skeletal projection fusion system for digital clothing refitting can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0138] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements:

[0139] Step 1: Obtain source data and target data; wherein, the source data includes a source digital character model and its bound source skeletal system and source clothing, and the target data includes a target digital character model and its bound target skeletal system; the source clothing has a three-dimensional geometric representation, including multiple vertices;

[0140] Step 2: Define local coordinate systems for each bone in the source and target skeletal systems, respectively;

[0141] Step 3: Based on the projection function, calculate the local projection coordinates of each vertex of the source clothing in the local coordinate system of each bone in the source skeleton system;

[0142] Step 4: Map the local projection coordinates to the semantic bone local coordinate system corresponding to the target bone system, and transform them to the global space through the inverse projection function to obtain the candidate global position set of each vertex for each bone in the target bone system;

[0143] Step 5: For each vertex, perform linear weighted fusion of the candidate global positions of its corresponding bones to determine the initial fitting position of the vertex in the target space, thereby obtaining the preliminary fitting target clothing.

[0144] Step 6: Based on the initial adaptation position, the final adaptation target clothing is generated by optimizing the coordinate residuals and fusion weight residuals in the local bone projection space.

[0145] For specific limitations on each step, please refer to the limitations on the local skeletal projection fusion method used for digital clothing refitting mentioned above, which will not be repeated here.

[0146] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program being implemented when executed by a processor:

[0147] Step 1: Obtain source data and target data; wherein, the source data includes a source digital character model and its bound source skeletal system and source clothing, and the target data includes a target digital character model and its bound target skeletal system; the source clothing has a three-dimensional geometric representation, including multiple vertices;

[0148] Step 2: Define local coordinate systems for each bone in the source and target skeletal systems, respectively;

[0149] Step 3: Based on the projection function, calculate the local projection coordinates of each vertex of the source clothing in the local coordinate system of each bone in the source skeleton system;

[0150] Step 4: Map the local projection coordinates to the semantic bone local coordinate system corresponding to the target bone system, and transform them to the global space through the inverse projection function to obtain the candidate global position set of each vertex for each bone in the target bone system;

[0151] Step 5: For each vertex, perform linear weighted fusion of the candidate global positions of its corresponding bones to determine the initial fitting position of the vertex in the target space, thereby obtaining the preliminary fitting target clothing.

[0152] Step 6: Based on the initial adaptation position, the final adaptation target clothing is generated by optimizing the coordinate residuals and fusion weight residuals in the local bone projection space.

[0153] For specific limitations on each step, please refer to the limitations on the local skeletal projection fusion method used for digital clothing refitting mentioned above, which will not be repeated here.

[0154] As a specific example, in one embodiment, the invention will be further described in detail with reference to the accompanying drawings.

[0155] This embodiment proposes a local skeletal projection fusion method to achieve digital clothing refitting, as shown in Figure 1, specifically including:

[0156] Step (1) Input source and target data: Obtain the source digital character model and the target digital character model through the Mixamo 3D character resource platform. The digital character model A includes triangle mesh vertex data, patch topology, and pre-bound skeleton system. and the corresponding skin weight information The acquired digital character model was imported into Blender 3D modeling software. Twenty bones and their corresponding skinning weights were extracted, and four auxiliary bones (left / right ribs and left / right hips) were added to construct a parent-child hierarchy. The skeletal system now includes multiple bone nodes and their parent-child hierarchical relationships. Each bone records its starting joint, ending joint, bone direction vector, and bone length information, ensuring consistency in topological structure and semantic correspondence between the source and target skeletal systems. A source-end clothing model was constructed in the Marvelous Designer 3D clothing modeling environment. Triangular mesh clothing data fitting the source character model was generated through 2D pattern modeling and physical simulation. The clothing mesh includes vertex coordinates, facet topology, and optional texture coordinate information, and maintains an initial pose consistent with the source digital character model. When verifying dynamic adaptation capabilities, character animation-driven data can be loaded, and corresponding clothing dynamic mesh sequence data can be generated through physical simulation. The aforementioned source and target data together constitute the input data basis of this method. Figure 3 shows a model wearing the source clothing. Source digital character model and its binding skeletal system A target digital character model and its binding skeletal system .

[0157] Step (2) is for the source skeletal system and target skeletal system Define a local coordinate system for each bone b: For any bone b, let its starting joint point be... The termination point is The bone direction vector is defined as follows: Bone length Normalize the bone direction vector and define the z-axis direction of the local coordinate system as... .

[0158] For the root skeleton, a reference vector in the global coordinate system is selected. The reference vector is... A fixed unit vector that is not collinear is used to orthogonalize the reference vector along the current z-axis direction to obtain the initial x-axis direction vector:

[0159]

[0160] The y-axis direction vector is then determined using the cross product operation:

[0161]

[0162] Thus, an initial local coordinate system satisfying the right-handed coordinate system constraint is constructed. .

[0163] For those with paternal skeletons The local coordinate system of bone b is constructed hierarchically through the coordinate system of its parent bone. Let the x-axis direction vector of the parent bone be... Then, the orthogonalization operation is used to remove its component in the current z-axis direction:

[0164]

[0165] And calculate the y-axis direction vector using the cross product:

[0166]

[0167] when When the x-axis direction of the parent bone is approximately parallel to the z-axis direction of the current bone, it is determined that... The preset threshold can be set to [value] in actual operation. At this point, the y-axis direction vector in the local coordinate system of the parent bone is selected. As a backup direction, the orthogonalization and normalization steps described above are repeated to ensure the numerical stability of the construction process.

[0168] Using the above recursive method, the source skeletal system With the target skeletal system Each bone in the structure is constructed with a consistent and stable local orthogonal coordinate system.

[0169] Step (3) For the source clothing 3D mesh model obtained in step (1) above It consists of a set of vertices. Composition, in which each vertex This represents the coordinate position of the source clothing in three-dimensional space, for each vertex. Using projection function Calculate the source skeletal system defined in step (2). middle bones Local projected coordinates in the local coordinate system: The source skeleton system has already been defined in step (2) above. Every bone A local orthogonal coordinate system was constructed. And determined the location of the starting joint of the bone. and bone length For any vertex First, calculate its displacement vector relative to the starting joint point of the bone. :

[0170]

[0171] Then, the displacement vector is projected onto the three coordinate axes of the bone's local coordinate system to obtain three scalar components:

[0172]

[0173] in, express and Dot product operation.

[0174] To eliminate the impact of differences in skeletal scale among different characters, the above three components are divided by the corresponding bone length. Normalization is performed to obtain scale-independent local projected coordinates:

[0175]

[0176] The above calculations are for the source skeletal system Every bone in All are performed, thus for each vertex Construct its multi-bone projection representation in different bone local coordinate systems.

[0177] Step (4) For the source vertex obtained in step (3) above In the source skeletal system Local projection coordinates of each bone Based on the semantic correspondence between the source skeleton and the target skeleton, in the target skeleton system Construct the inverse projection function corresponding to the forward projection function described in step (3). This is used to restore local projected coordinates to global spatial coordinates driven by the target bone: for the target bone Its starting joint position is The length of the skeleton is The local orthogonal basis vectors are The inverse projection function Defined as:

[0178] First, a scale restoration operation is performed, multiplying the normalized local projection coordinates from the forward projection phase by the target bone length. :

[0179]

[0180] Then, the restored local projected coordinates are linearly combined under the local orthogonal basis of the target bone to generate a displacement vector relative to the origin of the bone:

[0181]

[0182] Finally, the displacement vector is superimposed on the starting joint position of the target bone to obtain the candidate global position of the vertex driven by the target bone:

[0183]

[0184] Wherein, the inverse projection function Forward projection function The inverse mapping form in the local coordinate system of the target skeleton. By performing the above inverse projection calculation on each bone in the target skeleton system, the vertices can be obtained. Set of candidate global locations within the target skeleton system:

[0185]

[0186] This candidate global location set provides input for subsequent fusion weight calculation and vertex location determination. Figure 2 shows a schematic diagram of the local skeleton projection fusion method.

[0187] Step (5) For the set of candidate global positions of vertices in the target skeleton system obtained in step (4) The unique initial fit position of the vertex is determined by multi-bone weighted fusion: for the source clothing vertex When inputting data in step (1), the source skeletal system A set of skin weights has been pre-associated in the middle. ,in Let represent the weight coefficient of the vertex affected by bone b, and satisfy:

[0188] The skinning weights of the source garment are directly inherited as the initial fusion weights under the target skeletal system:

[0189]

[0190] The source skeleton system and the target skeleton system use the same semantic skeleton index set B, that is, each index b also corresponds to the source skeleton. With target skeleton Then, for each candidate global position... With corresponding fusion weights Multiplying these together yields the weighted contribution vector of each bone to that vertex. Finally, a linear weighted sum of all bone contribution vectors is performed to obtain the initial fitting position of the vertex in the target space:

[0191]

[0192] Through the above fusion operation, the initial geometric model of the target garment can be obtained. Figure 4 shows an initial target clothing result.

[0193] Step (6) Using the initial adaptive target clothing vertex positions obtained in step (5) As the initial state for optimization, the geometry of the target clothing is iteratively refined: the optimization variable is defined as the local coordinate residual of each vertex in the local coordinate system of the corresponding target skeleton. and fusion weight residual and initialized to .

[0194] For each vertex The local projected coordinates obtained in step (3) The local coordinates after the perturbation are:

[0195]

[0196] The perturbation is still defined in the local orthogonal coordinate system of the target skeleton.

[0197] To ensure the numerical stability of the weight adjustment, unconstrained variables are introduced. Through bounded mapping functions:

[0198]

[0199] Calculate the weighted residuals, where This is a preset scaling factor used to limit the magnitude of weight perturbation. In practice, it can be set to... Update the fusion weights to... And after normalization, we get:

[0200]

[0201] To meet .

[0202] Then, based on the perturbed local coordinates and the updated fusion weights, the inverse projection function is used. And the linear fusion structure calculates the current iteration vertex position:

[0203]

[0204] Through the aforementioned local residual perturbation and fusion weight update, the updated position of the current target clothing vertex can be calculated in one iteration. To ensure that the residual variables of each vertex converge stably under global constraints, an overall iterative optimization framework is constructed on this basis, and a coarse-to-fine hierarchical optimization strategy and a dynamic update mechanism for contact relationships are used for joint solution.

[0205] Within the aforementioned iterative optimization framework, a coarse-to-fine hierarchical optimization strategy is employed for iterative updates. and Specifically, this includes: firstly, analyzing the source clothing mesh. Downsampling is performed to generate a low-resolution mesh, and then a loss function is applied at this low-resolution level. Optimization is performed to obtain a coarse-resolution garment that initially matches the target character model; then, the coarse-resolution result is upsampled to restore a high-resolution mesh, and further optimization is carried out at the high-resolution level based on the loss function. The process involves refinement and optimization to restore the detailed features of the original clothing while ensuring convergence stability.

[0206] During the optimization process, a nearest neighbor search is performed every 2000 iterations to dynamically update the contact correspondence between the clothing mesh and the target character model. Specifically, in the initialization phase, the nearest bone segment is selected as the guide bone for each target clothing vertex. A ray is projected along the orthogonal direction of this bone segment and its intersection with the target character model is calculated to determine the corresponding initial contact point. After the clothing vertex converges to the correct surface neighborhood, the standard nearest neighbor query method is switched to update the contact pairs, thereby ensuring the accuracy and computational efficiency of the contact relationship and minimizing the impact on the loss function. It gradually converges to its minimum value. Loss function It is expressed as follows:

[0207]

[0208] in, The coefficients for the constraint terms can be set to [value] in practice. , This represents a constraint function that controls the penetration between clothing and characters and applies user-specified clothing fit effects.

[0209] In practice, Includes distance constraints and fit constraints Two parts:

[0210]

[0211] in It is a coefficient that controls the fit and style matching strength of clothing to the body. In practice, it can be set to... ;

[0212] Used to punish when clothing exceeds a preset threshold between the body and the clothing. The interleaving phenomenon is represented as follows:

[0213]

[0214] in, This represents the current vertex on the target garment. Indicates the distance from the target digital character model The nearest point, The normalized weights are related to the number of points. The target human body mesh area is proportional to the target human body mesh area. It is a distance threshold. express arrive Signed distance:

[0215]

[0216] in, Point The external normal vector of the human body surface at that location;

[0217] By using the user-specified fit control area The inner matching marked distance maintains the consistency of the garment's fit and style, as shown below:

[0218]

[0219] in, The signed distance from the vertex of the target clothing to the target digital character model. The signed distance from the source clothing vertex to the source digital character model, region This specifies the area where the refitted garment should inherit the style from the original garment. The system is constructed based on the z-axis coordinates of clothing vertices in a local skeletal coordinate system. Using the local skeletal system of the hip and spine as the criterion, clothing vertices are filtered by setting the sign or interval conditions of the z-axis coordinates, thus forming a fit control area. This area can be set as the upper torso region, waist region, a combination of upper torso and waist region, or a global region based on semantic input. During optimization, only vertices located within this region are optimized. The vertex within the model is used to calculate the fitting loss term. By matching the signed distance between the source garment and the target garment relative to the digital character model, the refitted garment maintains a fitting style consistent with the source garment within the specified area.

[0220] This represents a constraint function that maintains consistency in the shape and fold details of clothing. In practice, Includes Laplace terms Bending constraint and curvature constraints Three parts:

[0221]

[0222] in, This is the weighting factor, which can be set to 0.5 in practice. To preserve high-frequency wrinkle details, it is represented as follows:

[0223]

[0224] in, and Represent the vertices in the source and target clothing, respectively. The normalized Laplacian coordinates are obtained by normalizing the Laplacian coordinates:

[0225]

[0226] Among them, Laplace coordinates and The calculation method is as follows:

[0227]

[0228] in, Represents vertices In the first-order neighborhood of the corresponding grid For pre-calculated cotangent weights;

[0229] By maintaining the dihedral angles between garment panels To improve overall smoothness and suppress unwanted sharp creases, the following is indicated:

[0230]

[0231] in, For adjacent facets in the source garment, For the corresponding facet in the target garment, To normalize the weights, they are shared by the source garments. The length is predetermined;

[0232] Preserving the garment's silhouette and style by matching the cosine of the angle between boundary edges is represented as follows:

[0233]

[0234] in, For adjacent boundary edges in the source garment, For the corresponding edge in the target clothing, To normalize the weights, the corresponding boundary segment lengths in the source garment are used. Configure the settings.

[0235] The constraint function that regularizes the optimization variables stabilizes the optimization process by changing the constraint residuals, as shown below:

[0236]

[0237] in, This is used to penalize large residual changes along the bone axis to maintain a bone attachment relationship consistent with the source garment. This is used to regularize the fusion weight residuals to prevent drastic weight fluctuations. These are weighting coefficients, which can be set in practice. .

[0238] Once the optimization converges, the final refined set of vertex coordinates for the target clothing will be output:

[0239]

[0240] This completes the clothing refitting process on the target digital character model, as shown in Figure 5, which depicts a target digital character model wearing adapted target clothing.

[0241] In summary, this invention constructs a local skeletal projection fusion representation, processing clothing geometry within the local coordinate space of the skeleton. Compared to direct optimization in the global vertex coordinate space, this approach improves parameterization conditions and enhances the stability of the optimization process. Introducing residual variables into the local skeletal coordinate space for iterative optimization helps achieve fine-tuning of local geometry while maintaining overall shape consistency. The linear hybrid structure of fusion weights and the bounded constraint mechanism of weight residuals help expand the reasonable solution space and smooth the optimization process. Combining contact constraints and shape preservation constraints, this invention enables clothing refitting between digital character models of different body shapes and poses, while preserving the geometric details and fit of the clothing.

[0242] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention without departing from its spirit and scope should be included within the protection scope of the present invention.

Claims

1. A method for local skeletal projection fusion for digital clothing refitting, characterized in that, The method includes the following steps: Step 1, acquiring source data and target data; wherein, the source data includes a source digital character model and its bound source skeletal system and source clothing, and the target data includes a target digital character model and its bound target skeletal system; the source clothing has a three-dimensional geometric representation, including multiple vertices; Step 2, defining local coordinate systems for each bone in the source skeletal system and the target skeletal system respectively; Step 3, calculating the local projection coordinates of each vertex of the source clothing in the local coordinate systems of each bone in the source skeletal system based on a projection function; Step 4, mapping the local projection coordinates to the semantic bone local coordinate system corresponding to the target skeletal system, and transforming them to the global space through an inverse projection function to obtain a set of candidate global positions for each vertex relative to each bone in the target skeletal system; Step 5, for each vertex, performing linear weighted fusion of the candidate global positions of its corresponding bones to determine the initial adaptation position of the vertex in the target space, thereby obtaining a preliminary adapted target clothing; Step 6, based on the initial adaptation position, generating the final adapted target clothing by optimizing the coordinate residuals and fusion weight residuals in the local bone projection space.

2. The local skeletal projection fusion method for digital clothing refitting according to claim 1, characterized in that, In step 1, the source skeletal system and the target skeletal system have similar topological structures. Specifically, if the topological structures are consistent, a one-to-one semantic relationship is established between the bones of the source skeletal system and the target skeletal system. If the topological structures are different, the skeletal structures are adjusted through preset pruning or completion operations until the one-to-one semantic relationship is established.

3. The local skeletal projection fusion method for digital clothing refitting according to claim 1, characterized in that, Step 2 involves constructing a local coordinate system as follows: Step 2-1, using the starting joint of the skeleton as the origin of the local coordinate system, normalize the direction vector from the starting joint to the ending joint and use it as the z-axis direction of the local coordinate system; Step 2-2, select a reference vector or parent skeleton basis vector, eliminate its z-axis component through orthogonalization and normalize it, and use it as the x-axis direction of the local coordinate system; Step 2-3, determine the y-axis direction through the cross product of the z-axis and x-axis, thereby constructing a local coordinate system that satisfies the right-hand orthogonal constraint.

4. The local skeletal projection fusion method for digital clothing refitting according to claim 3, characterized in that, Step 2-2 specifically includes: when the bone is the root bone, a reference vector that is not collinear with the z-axis direction is selected, the reference vector is orthogonalized to eliminate its component in the z-axis direction, and after normalization, it is used as the x-axis direction; when the bone has a parent bone, a basis vector in the local coordinate system of the parent bone is selected as the initial direction, orthogonalized and normalized to obtain the x-axis direction of the current bone; wherein, when the initial direction is approximately parallel to the z-axis direction of the current bone, another basis vector in the local coordinate system of the parent bone is selected as the backup direction and the orthogonalization process is re-executed.

5. The local skeletal projection fusion method for digital clothing refitting according to claim 1, characterized in that, Step 3 specifically includes: Step 3-1, for each vertex of the source garment, calculate its displacement vector relative to the starting joint point of the skeleton; Step 3-2, based on the projection function, project the displacement vector onto the three orthogonal basis vector directions of the local coordinate system of the skeleton, and obtain the three directional components through dot product operation; Step 3-3, divide the three directional components by the corresponding bone length for normalization processing to construct scale-independent local projection coordinates; Steps 3-1 to 3-3 are performed on each bone in the source skeleton system to construct the projection representation of the vertex in the multi-bone local coordinate space.

6. The local skeletal projection fusion method for digital clothing refitting according to claim 1, characterized in that, Step 4 specifically includes: Step 4-1, scaling the local projected coordinates using the target bone length; the target bone is the bone in the target bone system corresponding to each bone in the source bone system; Step 4-2, rotating and synthesizing the scaled coordinates using the basis vectors of the target bone to generate a displacement vector relative to the bone origin; Step 4-3, superimposing the displacement vector onto the starting joint of the target bone to obtain the candidate global position of the vertex relative to the target bone; Steps 4-1 to 4-3 are performed on each bone in the target bone system to obtain the set of candidate global positions of the vertex relative to each bone in the target bone system.

7. The local skeletal projection fusion method for digital clothing refitting according to claim 1, characterized in that, Step 5 determines the initial adaptation position. The specific process includes: Step 5-1, for each vertex of the source clothing, a set of skinning weights pre-associated with it is inherited as a set of initial fusion weights under the target skeleton system; where it is necessary to ensure that all associated bones of the vertex correspond to an initial fusion weight; Step 5-2, for each vertex, the candidate global position of each bone obtained in Step 4 is multiplied by the corresponding initial fusion weight and summed to obtain the initial adaptation position of the vertex in the target space; where, during the fusion process, the sum of the initial fusion weights of all associated bones of each vertex must be 1.

8. The local skeletal projection fusion method for digital clothing refitting according to claim 7, characterized in that, Step 6 specifically includes the following steps: Step 6-1, for each vertex, add its local projected coordinates in the local coordinate system of each bone obtained in Step 3 to the local coordinate residuals of the vertex relative to each bone to obtain the perturbed local coordinates in the semantic bone local coordinate system; Step 6-2, introduce unconstrained variables and combine them with bounded mapping functions to calculate the fusion weight residuals; Step 6-3, for each initial fusion weight, sum it with the fusion weight residuals and normalize it to obtain the updated fusion weights; Step 6-4, based on the perturbed local coordinates and the updated fusion weights, combine them with the target loss function for iterative optimization to obtain the updated positions of each vertex of the initially adapted target clothing.

9. The local skeletal projection fusion method for digital clothing refitting according to claim 8, characterized in that, The objective loss function includes contact constraints. Shape retention constraints and regularization terms , is represented as: In the formula, Represents the target loss function. This represents a constraint function that controls the penetration between clothing and characters and applies user-specified clothing fit effects. This represents a constraint function that maintains consistency in the shape and fold details of the garment. This represents a constraint function that performs regularization on the optimization variables. These are the coefficients of the constraint terms.

10. The local skeletal projection fusion method for digital clothing refitting according to claim 9, characterized in that, In step 6, a coarse-to-fine hierarchical optimization strategy and a dynamic update mechanism for contact relationships are adopted to optimize the target loss function. The optimization process gradually converges to a minimum value. Specifically, the coarse-to-fine hierarchical optimization strategy includes: performing initial matching optimization on a downsampled low-resolution mesh; upsampling the initial matching optimization result to a high-resolution mesh and performing fine-tuning optimization based on the target loss function to obtain a high-resolution garment that matches the target digital character model. The dynamic update mechanism for contact relationships specifically includes: selecting the nearest bone segment as a guide bone for each vertex of the initially adapted target garment; projecting rays along the orthogonal direction of the bone segment and intersecting with the target digital character model to determine the initial contact point; and after the vertices of the initially adapted target garment converge to the surface neighborhood of the target digital character model, switching to a nearest neighbor search method to update the contact correspondence between the garment and the target digital character model.

Images