A 3D Motion Posture Capture Data Restoration Method, System, Device, and Medium

Abstract

The present invention is applicable to the field of computer vision technology, and provides a method, a system, a device and a medium for restoring 3D motion posture capture data. The method for restoring 3D motion posture capture data includes the following steps: obtaining 3D human motion data M; performing multiple fine-grained divisions on the 3D human motion data according to the correlation between different parts of the human body to obtain a plurality of fine-grained modules; using a low-rank regularization term to regularize each of the fine-grained modules in a data restoration model; introducing two types of low-rank progressive norms, convex and non-convex, into the data restoration model; calculating and solving the data restoration model to obtain a restoration result. This method strengthens the low-rank characteristics of the data, reduces the restoration error, and obtains a more superior restoration result.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer vision, and particularly relates to a method, system, device and medium for restoring 3D motion posture capture data. Background Art

[0002] Currently, the methods for restoring 3D motion posture capture data can be roughly divided into four categories: signal processing-based, interpolation-based, data-driven, and low-rank matrix completion (LRMC). Traditional signal processing-based methods regard motion data as ordinary signals and use Gaussian low-pass filters or Kalman filters to denoise the data. Since each degree of freedom is processed independently, the spatio-temporal characteristics of the motion data are destroyed, and the restored results often have obvious visual defects. Interpolation-based methods mainly utilize the temporal continuity of human motion and fill in the missing data points by averaging the data values on both sides of the missing points. It has the characteristics of low time cost and easy implementation. However, such methods ignore the structural correlation between different joints and are often helpless for long-term missing data.

[0003] With the explosive growth of existing motion data, researchers have turned to using data-driven strategies to fill in the defective 3D human motion data. However, data-driven methods require a clean data set as support. Due to the randomness of the missing motion data and the continuous increase in the types of motion data over time, it is too time-consuming and laborious to establish a large data set that can continuously contain all motion data. Therefore, methods based on low-rank matrix completion (LRMC) have been favored by researchers. However, most LRMC-based methods treat all motion parts of the human body equally, discrete the correlation between the same human body parts, and violate the low-rank assumption of the data, resulting in suboptimal restoration results. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide a method for restoring 3D motion posture capture data, aiming to strengthen the low-rank characteristics of the data, reduce the restoration error, and obtain a more superior restoration result.

[0005] The embodiments of the present invention are implemented as follows. A method for restoring 3D motion posture capture data, the method for restoring 3D motion posture capture data includes the following steps:

[0006] Obtain 3D human motion data M;

[0007] According to the correlation between different parts of the human body, perform multiple fine-grained partitions on the 3D human motion data to obtain a number of fine-grained modules;

[0008] Use a low-rank regularization term to regularize each of the fine-grained modules in the data restoration model;

[0009] Two types of convex and non - convex low - rank progressive norms are introduced into the data restoration model;

[0010] Solve the data restoration model to obtain the restoration result.

[0011] Furthermore, the multi - fine - grained partitioning of the 3D human motion data specifically includes the following steps:

[0012] Set fine - grained parameters according to the correlation of different parts of the human body;

[0013] Use the following formula for multi - fine - grained partitioning:

[0014] ;

[0015] where X is the data reconstructed from the 3D human motion data M; is the fine - grained partitioning operator; d represents the number of modules after partitioning, that is, the fine - grained parameter; c represents the c - th module under different fine - grained partitioning conditions.

[0016] Furthermore, the construction method of the data restoration model includes the following steps:

[0017] Merge the fine - grained modules of the same scale into one model, introduce a low - rank regularization term to regularize each fine - grained module, and obtain the data restoration model:

[0018] ;

[0019] where M represents the 3D human motion data; X represents the data reconstructed from the 3D human motion data M; Ω is the observation matrix; represents the Hadamard product operation of matrices; represents the c - th module after the d - th level of fine - grained partitioning; is the low - rank regularization term; λ is the regularization parameter.

[0020] Furthermore, the following optimizations are performed on the data restoration model:

[0021] ;

[0022] ;

[0023] where is the time - stability term after calculating the difference matrix; is a tridiagonal matrix, and each column of the matrix XO calculates the difference between adjacent frames in X; is the Frobenious norm.

[0024] Furthermore, the convex and non-convex low-rank progressive norms are specifically as follows:

[0025] Convex relaxation of the rank function: Its nuclear norm is:

[0026] ;

[0027] Non-convex relaxation of the rank function: The Schatten p-norm is:

[0028] ;

[0029] Wherein, represents the i-th singular value of matrix X, p represents the cumulative measure of taking the p-th power of the singular values of the matrix, and the value range of p is 0-1.

[0030] Furthermore, the calculation and solution of the data restoration model are specifically as follows:

[0031] Introduce auxiliary variables A = X and S in the data restoration model c = AD c ;

[0032] Establish the augmented Lagrangian function of the data restoration model;

[0033] Construct a sub-problem regarding the auxiliary variables and solve it under the framework of the alternating direction multiplier to obtain the restoration result.

[0034] Furthermore, after calculating and solving the data restoration model to obtain the restoration result, it further includes:

[0035] Fuse each restoration result calculated by the data restoration model to obtain the restored 3D human motion data.

[0036] Another object of the embodiment of the present invention is a 3D motion pose capture data restoration system, which is characterized in that the 3D motion pose capture data restoration system executes the 3D motion pose capture data restoration method, and the system includes:

[0037] An acquisition module, configured to acquire 3D human motion data M;

[0038] A partitioning module, which performs multiple fine-grained partitions on the 3D human motion data according to the correlation between different parts of the human body to obtain several fine-grained modules;

[0039] A 3D human motion restoration module, which uses a low-rank regularization term to regularize each of the fine-grained modules in the data restoration model; introduces convex and non-convex low-rank progressive norms in the data restoration model; calculates and solves the data restoration model to obtain the restoration result.

[0040] Another object of an embodiment of the present invention is a computer device, characterized by comprising a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the 3D motion pose capture data restoration method.

[0041] Another object of an embodiment of the present invention is a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the processor executes the steps of the 3D motion pose capture data restoration method.

[0042] The 3D motion pose capture data restoration method provided by an embodiment of the present invention treats the human body data of the same part as a general unit, so that the restoration work focuses on the data with strong correlations. In order to combine the characteristics of the data after different fine-grained partitions, the present invention also performs a fusion process on the restoration results after fine-grained partitioning using different parameters. Since the data of the same part has more significant low-rank characteristics, compared with the LRMC method that equally processes all motion parts, the multiple fine-grained fusion method can strengthen the low-rank characteristics of the data, reduce the restoration error, and obtain a more superior restoration result. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is an application environment diagram of the 3D motion pose capture data restoration method provided by an embodiment of the present invention;

[0044] Figure 2 is a flowchart of the 3D motion pose capture data restoration method provided by the first embodiment;

[0045] Figure 3 is a flowchart of the 3D motion pose capture data restoration method provided by the first embodiment;

[0046] Figure 4 is an internal structure block diagram of the computer device in the third embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0048] Figure 1 is an application environment diagram of the 3D motion pose capture data restoration method provided by an embodiment of the present invention, as Figure 1 shown. In this application environment, an image acquisition device and a computer device are included.

[0049] An image acquisition device is a tool used to capture and convert visual information (such as photos or videos) into digital format so that a computer can process this data. It can be a scanner, a camera, an industrial camera, a vision sensor, etc.

[0050] A computer device is a broad term that encompasses various hardware used for input, processing, output, and storage of data. It can be a laptop or a desktop computer, an independent physical server or terminal, or a server cluster composed of multiple physical servers. The computer device obtains 3D human motion data from the image acquisition device.

[0051] As Figure 2 shown, in the first embodiment, a method for restoring 3D motion pose capture data is proposed. This embodiment takes the application of this method to the Figure 1 above-mentioned computer device as an example. A method for restoring 3D motion pose capture data may specifically include the following steps S102 to S112:

[0052] Step S102, obtain 3D human motion data M.

[0053] In step S102, the 3D human motion data M is used to record the three-dimensional position and orientation information of the moving body. Before performing restoration processing on the 3D human motion data extracted from the three-dimensional space using the LRMC-based method, the motion needs to be organized frame by frame in a matrix manner, and each frame records the three-dimensional position of each joint moment, which can be specifically formulated as:

[0054] ;

[0055] ;

[0056] Among them, is the motion data matrix; is the maker position of the t-th frame, 1 ≤ t ≤ n; m is the number of joints in a frame; n is the number of frames.

[0057] Step S104, according to the correlation between different parts of the human body, perform multiple fine-grained partitions on the 3D human motion data to obtain several fine-grained modules.

[0058] In step S104, the Mocap data processed by constructing a fine-grained operator is used as a general unit to effectively utilize the correlation of the same human body part. Specifically, step S104 can be refined into steps S202 to S204:

[0059] Step S202, set fine-grained parameters according to the correlation between different parts of the human body;

[0060] Step S204, perform multi - level fine - grained partitioning using the following formula:

[0061] ;

[0062] where X is the data reconstructed from the 3D human motion data M; is the fine - grained partitioning operator; d represents the number of modules after partitioning, i.e., the fine - grained parameter; c represents the c - th module under different fine - grained partitioning conditions.

[0063] In steps S202 - S204, since it is difficult to determine the size of the partitioned parts and the fine - grained parameter, if the fine - grained parameter is too small or too large, it is impossible to well explore the strongly related parts of the human body. Therefore, to improve the recovery accuracy, this embodiment proposes to use the multi - level fine - grained partitioning method. Specifically, according to the correlation between different parts of the human body, the fine - grained parameters are set to d = 2, d = 3, and d = 5, which can divide the human body into two upper and lower parts; three parts including the upper body and two legs; five parts including the limbs and the torso respectively.

[0064] Step S106, use the low - rank regularization term to regularize each of the fine - grained modules in the data restoration model.

[0065] In step S106, in the practical process, considering that each module after fine - grained partitioning is affected by its surrounding environment, and finally the restored data needs to be globally smoothed, this embodiment combines the fine - grained modules of the same scale into one model instead of evaluating them separately, and introduces the low - rank regularization term to regularize each fine - grained module. Specifically, the construction method of the data restoration model in step S106 is as follows:

[0066] ;

[0067] where M represents the acquired 3D human motion data; X represents the data reconstructed from the 3D human motion data M; Ω is the observation matrix, when the motion data is missing or noisy, the element value in Ω is 0, when the motion data is observed, the element value in Ω is 1; represents the Hadamard product operation of matrices; represents the c - th module after the d - th level of fine - grained partitioning; is the low - rank regularization term; λ is the regularization parameter.

[0068] Furthermore, considering that the 3D human motion data has a temporal structure, to improve the temporal smoothness, the data restoration model can be further optimized, and the optimization method is:

[0069] ;

[0070] ;

[0071] wherein, is the time stability term after differential matrix calculation; is a tridiagonal matrix, and each column of matrix XO calculates the difference between adjacent frames in X; is the Frobenious norm. Therefore, by selecting an appropriate λ, the objective function minimizes the total difference between adjacent frames, effectively avoiding unevenness between adjacent frames.

[0072] Step S108, introduce two types of convex and non-convex low-rank progressive norms into the data restoration model.

[0073] In step S108, since the rank minimization problem is an NP-hard problem, in this embodiment, two types of convex and non-convex low-rank progressive norms will be introduced into the established model respectively. By flexibly introducing different low-rank norms, the low-rank characteristics of the fine-grained module are utilized to further reduce the error in the process of 3D human motion data restoration. Specifically as follows:

[0074] Convex relaxation of the rank function: Its nuclear norm is:

[0075] ;

[0076] Non-convex relaxation of the rank function: The Schatten p-norm is:

[0077] ;

[0078] wherein, represents the i-th singular value of matrix X, p represents the cumulative metric of taking the p-th power of the singular values of the matrix, and the value range of p is 0 - 1.

[0079] In step S108, since the data of the same part has more significant low-rank characteristics, compared with the LRMC method that equally processes all motion parts, the multi-fine-grained fusion method can strengthen the low-rank characteristics of the data, reduce the restoration error, and obtain a more superior restoration result.

[0080] Step S110, calculate and solve the data restoration model to obtain the restoration result.

[0081] In step S110, since the same low-rank progressive method is adopted for the 3D human data divided by different fine-grained parameters, the subsequent calculation can omit the explicit representation method of the modules divided at different levels. Calculating and solving the data restoration model specifically includes steps S302 - 306:

[0082] Step S302, introduce auxiliary variables \(A = X\) and \(S\) into the data restoration model c \(= AD\) c ;

[0083] Step S304, establish the augmented Lagrangian function of the data restoration model;

[0084] Step S306, construct a sub - problem regarding the auxiliary variable and solve it under the framework of the Alternating Direction Method of Multipliers (ADMM) to obtain the restoration result.

[0085] The data restoration model in step S302 is the same as that in step S106. The augmented Lagrangian function in step S304 is as follows:

[0086]

[0087] where, \(Z1\), \(Z2\) and are Lagrange multipliers; \(\mu>0\), \(\beta>0\) and \(\rho>0\) are penalty parameters. Under the ADMM framework, in this embodiment, the restored data \(X\) can be accurately obtained through the following iterative scheme:

[0088]

[0089] The above problem involves 3 sub - problems: the sub - problem of \(S\) c the sub - problem of \(A\) and the sub - problem of \(X\).

[0090] 1. The sub - problem regarding \(S\) c :

[0091] ;

[0092] where, \(R(\cdot)\) represents different low - rank regularization terms, which can be the nuclear norm or the Schatten p - norm. When the introduced low - rank regularization term is the nuclear norm, the closed - form solution of \(S\) c can be obtained according to the singular value shrinkage operator:

[0093] ;

[0094] where, is the singular value shrinkage operator, defined as follows:

[0095] ;

[0096] where, \(X = U\sum V\) T is the singular value decomposition of \(X\); \(\sum=diag(\sigma\) i ), and \(1\leq i\leq min\{m,n\}\); \(D\) δ(∑) = diag(max{σ i - δ}, 0); X is an m×n matrix, and there exists a decomposition such that U is an m×m matrix; Σ is a positive semi - definite m×n diagonal matrix; and V T that is, V, is an n×n matrix; The singular value decomposition operation of X can be directly obtained through the command [U,Σ,V]=svd(X) of matlab software.

[0097] When the introduced low - rank regularization term is the Schatten p - norm, the re - weighted method can be used to solve it. According to the first - order optimality condition, by taking the derivative of S c and setting it to 0, we can get:

[0098] ;

[0099] where, is the derivative of the Schatten p - norm of S c with respect to S c and can be regarded as a re - weighted term in the above formula; I n represents an n×n identity matrix, Y c represents the updated Lagrange multiplier, which can be seen when performing the ADMM framework. The superscript t + 1 in the ADMM framework represents the (t + 1) - th iteration.

[0100] 2. Sub - problem about A:

[0101] .

[0102] According to the first - order optimality condition, by taking the derivative of A and setting it to 0, the optimal solution can be obtained:

[0103] .

[0104] 3. Sub - problem about X:

[0105] .

[0106] Since the Hadamard product is involved in the solution process, after putting the matrix elements in a certain state, according to the first - order optimality condition, by taking the derivative of X in the formula and setting it to 0, the optimal solution can be obtained:

[0107] .

[0108] Step S112, fuse each restoration result calculated by the data restoration model to obtain the restored 3D human motion data.

[0109] In step S112, in order to combine the features of all recovery results under different fine-grained parameters, this embodiment will perform average fusion on the recovery results of all levels after recovering data from each level partition in a set of fine-grained levels.

[0110] like Figure 3 As shown, in one example, the fine-grained parameters are set to 2, 3 and 5 respectively, and the fine-grained modules A, B and C are obtained after the fine-grained division. Two different low-rank regularization norms are introduced to obtain three restoration results, and the restoration results are fused to obtain the restored 3D human motion data X. It should be noted that the fine-grained parameters of this embodiment can be set according to actual conditions and are not limited to 2, 3 and 5. The 3D motion posture capture data restoration method proposed in this embodiment proposes multiple fine-grained division operators on the basis of the correlation between the same parts of the human body, and focuses the restoration of 3D human motion on strongly correlated data. By introducing two different low-rank regularization norms, the local low rank of the data is more effectively utilized. In order to stabilize the global features of the restored data after division with different fine-grained parameters, this embodiment performs fusion processing on the restored data under all fine-grained parameters, which improves the problem of too few features and unstable restoration results caused by the previous use of a single fine-grained division. In addition, after fine-grained division, data can theoretically be restored through parallel processing. For single-semantic long sequence data, this can effectively save data recovery time and improve overall recovery efficiency.

[0111] In a second embodiment, a 3D motion gesture capture data restoration system is provided. The 3D motion gesture capture data restoration system can be integrated into Figure 1 The computer device may specifically include:

[0112] An acquisition module, used for acquiring 3D human motion data M;

[0113] A division module, which performs multiple fine-grained divisions on the 3D human motion data according to the correlation between different parts of the human body to obtain a plurality of fine-grained modules;

[0114] The 3D human motion restoration module utilizes a low-rank regularization term to regularize each of the fine-grained modules in the data restoration model; introduces two low-rank asymptotic norms, convex and non-convex, into the data restoration model; and calculates and solves the data restoration model to obtain a restoration result.

[0115] In the second embodiment, the 3D motion posture capture data restoration system can execute any steps and sub-steps in the first embodiment, which will not be described in detail in this embodiment.

[0116] Figure 4Shows the internal structure diagram of the computer device in the third embodiment. The computer device may specifically be the Figure 1 computer device in. As Figure 4 shown, the computer device includes a processor, a memory, a network interface, and an input device connected through a system bus. Among them, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor can implement the 3D motion pose capture data restoration method. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can execute the 3D motion pose capture data restoration method. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device may be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the computer device housing, or an external keyboard, touchpad, or mouse, etc.

[0117] Those skilled in the art can understand that Figure 4 the structure shown in is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0118] In the fourth embodiment, a computer device is proposed. The computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, any step or sub-step of the first embodiment is implemented, such as the following steps:

[0119] Step S102, obtaining 3D human motion data M;

[0120] Step S104, according to the correlation between different parts of the human body, performing multiple fine-grained partitions on the 3D human motion data to obtain several fine-grained modules;

[0121] Step S106, using a low-rank regularization term to regularize each of the fine-grained modules in the data restoration model;

[0122] Step S108, introducing two types of convex and non-convex low-rank progressive norms into the data restoration model;

[0123] Step S110, calculating and solving the data restoration model to obtain a restoration result.

[0124] In a fifth embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, any step or sub-step of the first embodiment is implemented, such as the following steps:

[0125] Step S102, obtaining 3D human motion data M;

[0126] Step S104, according to the correlation between different parts of the human body, performing multiple fine-grained partitions on the 3D human motion data to obtain a number of fine-grained modules;

[0127] Step S106, using a low-rank regularization term to regularize each of the fine-grained modules in the data restoration model;

[0128] Step S108, introducing two types of low-rank progressive norms, convex and non-convex, into the data restoration model;

[0129] Step S110, calculating and solving the data restoration model to obtain a restoration result.

[0130] It should be understood that although the steps in the flowcharts of the embodiments of the present invention are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in each embodiment may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least a part of other steps or sub-steps or stages of other steps.

[0131] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0132] The above embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the appended claims.

[0133] The foregoing is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for restoring 3D motion posture capture data, characterized in that, The 3D motion pose capture data restoration method includes the following steps: Obtain 3D human motion data M; According to the correlation between different parts of the human body, perform multiple fine-grained partitions on the 3D human motion data to obtain a number of fine-grained modules; Using a low-rank regularization term, regularize each of the fine-grained modules in the data restoration model; Introduce two types of convex and non-convex low-rank progressive norms into the data restoration model; Calculate and solve the data restoration model to obtain a restoration result; The construction method of the data restoration model includes the following steps: Merge the fine-grained modules of the same scale into one model, and introduce a low-rank regularization term to regularize each of the fine-grained modules to obtain the data restoration model: ； Among them, M represents the 3D human motion data; X represents the data after reconstructing the 3D human motion data M; Ω is the observation matrix; represents the Hadamard product operation of matrices; represents the c-th module after the d-th level of fine-grained division; is the low-rank regularization term; λ is the regularization parameter; The multiple fine-grained partitioning of the 3D human motion data specifically includes the following steps: Set fine-grained parameters according to the correlation between different parts of the human body; Use the following formula for multiple fine-grained partitioning: ； Among them, is a fine-grained partitioning operator; d represents the number of modules after partitioning, that is, the fine-grained parameter; c represents the c-th module under different fine-grained partitioning conditions; Perform the following optimization on the data restoration model: ； ； Among them, is the temporal stability term after differential matrix calculation; is a tridiagonal matrix, and each column of matrix XO calculates the difference between adjacent frames in X; is the Frobenious norm; The two types of convex and non-convex low-rank progressive norms are specifically as follows: Convex relaxation of the rank function, whose nuclear norm is: ； Non-convex relaxation of the rank function, and the Schatten p-norm is: ； Among them, represents the i-th singular value of matrix X, p represents the cumulative measure of taking the p-th power of the singular values of the matrix, and the value range of p is 0-1.

2. The 3D motion attitude capture data restoration method according to claim 1, wherein Calculate and solve the data restoration model, specifically as follows: Introduce auxiliary variables A = X and S into the data restoration model c = AD c ; Establish an augmented Lagrangian function of the data restoration model; Construct a sub-problem regarding the auxiliary variable and solve it in the framework of the alternating direction multiplier to obtain the restoration result.

3. The 3D motion attitude capture data restoration method according to claim 1, characterized in that, After calculating and solving the data restoration model to obtain a restoration result, it further includes: Fuse each restoration result calculated by the data restoration model to obtain the restored 3D human motion data.

4. A 3D motion posture capture data restoration system, characterized in that, The 3D motion pose capture data restoration system executes the 3D motion pose capture data restoration method according to any one of claims 1-3. The system includes: An acquisition module for acquiring 3D human motion data M; A partitioning module that, according to the correlation between different parts of the human body, performs multiple fine-grained partitions on the 3D human motion data to obtain a number of fine-grained modules; A 3D human motion restoration module that uses a low-rank regularization term to regularize each of the fine-grained modules in the data restoration model; introduces two types of convex and non-convex low-rank progressive norms into the data restoration model; calculates and solves the data restoration model to obtain a restoration result.

5. A computer device, characterized in that, It includes a memory and a processor. A computer program is stored in the memory. When the computer program is executed by the processor, the processor executes the steps of the 3D motion pose capture data restoration method according to any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium. When the computer program is executed by the processor, the processor executes the steps of the 3D motion pose capture data restoration method according to any one of claims 1 to 3.

Images