A method for correcting noise and errors in human motion capture data

Abstract

The present invention belongs to the field of data correction technology and discloses a method for correcting noise and errors in human motion capture data. The present invention processes human motion capture data; a method for retrieving human motion capture data constructs a VGG network with a hash layer added before the classification layer; by simultaneously minimizing a loss function defined by the classification error and constraints on the hash code, the learned compact binary hash code can achieve good discrimination capabilities and a balanced bit distribution, effectively improving the retrieval accuracy and efficiency of human motion capture data; and, by performing animation synthesis and editing on the human motion capture data, a generative model is constructed, providing an intuitive and convenient interactive means for non-professional animators to use the model to generate high-quality animations.

Description

Technical Field

[0001] The present invention belongs to the technical field of data correction, and in particular relates to a method for correcting noise and errors in human motion capture data. Background Art

[0002] A motion capture system is a high-tech device used to accurately measure the motion of moving objects in three-dimensional space. Based on computer graphics principles, it uses multiple video capture devices arranged in space to record the motion of moving objects (trackers) as images. Computers then process this image data to obtain the spatial coordinates (X, Y, Z) of different objects (trackers) at different time units. However, existing methods for correcting noise and errors in human motion capture data have low data retrieval efficiency and poor animation quality for human motion synthesis.

[0003] Through the above analysis, the problems and defects of the existing technology are as follows:

[0004] (1) Existing methods for correcting noise and errors in human motion capture data have low data retrieval efficiency.

[0005] (2) The synthesis effect of human motion animation is poor. Summary of the Invention

[0006] In view of the problems existing in the prior art, the present invention provides a method for correcting noise and errors in human motion capture data.

[0007] The present invention is achieved by: a method for correcting noise and errors in human motion capture data comprising:

[0008] Step 1: Arrange cameras in multiple directions around the human body to collect human body motion data through the cameras; construct a human body geometric model through a modeling program; set human body mass distribution parameters; and input the captured human body motion data;

[0009] The method for collecting human motion data by using a camera:

[0010] The human body wears a cursor that reflects a specific infrared laser, and multiple infrared laser cameras are deployed around the scene;

[0011] The laser camera will actively emit specific infrared lasers towards the human body;

[0012] When the laser hits the cursor, it reflects; the camera calculates the spatial position of each cursor by receiving these reflected lights and obtains human motion data;

[0013] The method for constructing a human body geometric model through a modeling program:

[0014] Acquire whole-body image scanning data of a human body; segment the whole-body image data of the human body to obtain human body segmentation data;

[0015] Performing three-dimensional reconstruction on the human body segmentation data to obtain a voxelized anisotropic human body geometric model; converting the voxelized anisotropic human body geometric model into a curved anisotropic human body geometric model;

[0016] Segmenting the whole-body image data to obtain the body segmentation data includes: performing structural segmentation on the whole-body image data according to the human anatomical structure, and identifying anisotropic tissues and organs with different directions in the whole-body image data;

[0017] Identifying the anisotropic tissues with different directions and the anisotropic organs with different directions as different sub-tissues and different sub-organs to obtain the segmentation data;

[0018] The converting of the voxelized anisotropic human body geometric model into a curved anisotropic human body geometric model comprises: applying a curved surface model generation algorithm to reconstruct various tissues and organs of the voxelized anisotropic human body geometric model into curved surface anisotropic human body geometric models; the curved surface model generation algorithm comprises a marching cubes algorithm;

[0019] Step 2: Process the human motion capture data; retrieve the human motion capture data; parameterize the human motion joints according to the motion data; establish a hybrid Newton-Euler multi-rigid body dynamics equation based on Euler angles and quaternions; establish an optimization model based on physical constraints; set the number of iterations of the optimization calculation; solve the optimization model using a stepwise quadratic programming method; and obtain human motion data that meets the physical constraints.

[0020] The method for processing human motion capture data:

[0021] Build a human body database and store the captured human body motion data into the human body database;

[0022] Calibrate human motion capture data and remove inaccurate data;

[0023] Perform statistics on human motion capture data through statistical programs; and analyze the human motion capture data;

[0024] Step three: Animation synthesis and editing of human motion capture data.

[0025] Furthermore, the human body motion capture data is processed; the method for retrieving the human body motion capture data is as follows:

[0026] (1) constructing a human motion capture database; generating a human motion representation image for each segment of human motion capture data based on a human motion capture data sequence in the human motion capture database; constructing a deep hash feature extraction model, and extracting hash features of the human motion representation image through a pre-trained deep hash feature extraction model;

[0027] The deep hash feature extraction model is based on the deep neural network VGG16, and a hash layer activated by a sigmoid function is added before the classification layer of the deep neural network VGG16. During the training process of the deep hash feature extraction model, the following constraints are added to the loss function:

[0028] (1)

[0029] (2) J3 = tr(Out(H)(Out(H)) T )

[0030] Where Out(H) is the feature vector output by the hash layer in the deep hash feature extraction model, and e represents a one-dimensional vector with all elements equal to 1;

[0031] (2) For the human motion capture data segment to be retrieved, extract its hash feature, and calculate the distance between the hash feature of the human motion capture data to be retrieved and the hash feature of each segment of human motion capture data in the human motion capture database, and select the first k data with the closest distance as the retrieval result.

[0032] Furthermore, the generation of the human body motion representation image comprises the following steps:

[0033] Acquire human motion capture data from a human motion capture database;

[0034] Based on forward human kinematics, the joint angle information in the human motion capture data is converted into the trajectory information of each joint point of the human body;

[0035] The spatial configuration and dynamic characteristics of the joint trajectory over a period of time are represented as three channels of RGB images, which are converted from the trajectory, velocity field and self-similarity of the joint respectively.

[0036] Furthermore, among the three channels of the RGB image, the first channel is encoded by the trajectory of the joints, which retains the spatial information of the joints in each frame; the second channel is encoded by the velocity field, which is composed of the velocity vector of each joint and describes the human kinematic characteristics of the human moving object; the third channel reflects the change in the distance between each joint pair, which is encoded by the self-similar matrix connecting the joints along the time dimension.

[0037] Furthermore, the hash features of the human motion representation image are extracted by using a pre-trained deep hash feature extraction model, and the specific steps are as follows:

[0038] Training the deep hash feature extraction model using human motion representation images;

[0039] The training termination condition is that the classification accuracy of the deep hash feature extraction model is greater than a preset threshold;

[0040] The features of the hash layer of the deep hash feature extraction model are extracted as hash features of the human motion representation image.

[0041] Furthermore, the method for performing animation synthesis and editing on the data captured by human motion is as follows:

[0042] 1) Perform simple preprocessing on human motion capture data to annotate the prior information required to construct the generative model; define the random process based on the annotated information, and assign a suitable kernel function to each factor based on the specific meaning of each factor to describe the mapping relationship between it and the human motion capture data, thereby modeling the human motion capture data;

[0043] 2) Construct an optimization objective function and use a numerical optimization algorithm to solve the unknown parameters to build a generative model; using the obtained generative model, the decomposed factors are edited to achieve the synthesis and editing of human motion.

[0044] Furthermore, the specific contents of the preprocessing and labeling of human motion capture data are as follows:

[0045] Step (A1): Calculate the feature vector based on the human motion capture data set. The feature vector includes the root node position offset, the root node orientation, the degrees of freedom of each joint point, and the acceleration of each joint;

[0046] Step (A2) labels each human motion segment, wherein the labeling content specifically includes: person, style, content, gender, and age information.

[0047] Furthermore, the steps of constructing the generative model are specifically as follows:

[0048] Step (B1), define a random process based on the information marked in step (1);

[0049] Step (B2): assigning a suitable kernel function to each factor in the random process defined in step (B1) according to the specific meaning of each factor. The kernel function can be any valid kernel function such as RBF function and linear function.

[0050] Step (B3), after specifying the kernel function for each factor, the covariance matrix of each factor is calculated, and then the covariance matrix of the generated model is obtained by multiplying the covariance matrices, and then the modeling of the human motion capture data is completed.

[0051] Furthermore, the steps of constructing the objective function to solve the unknown parameters are as follows:

[0052] Step (C1), constructing a conditional probability formula and constructing the objective function using maximum a posteriori estimation (MAP);

[0053] Step (C2): using the numerical optimization algorithm SCG to solve the unknown parameters to obtain the generative model.

[0054] Furthermore, the specific content of using the generative model to realize human motion synthesis and editing is as follows:

[0055] Content (D1), synthesize new human motion by specifying factor parameters;

[0056] Content (D2), by estimating the unknown human motion parameters, the human motion and human motion style are oriented;

[0057] Content (D3), realizes the fusion of two or more human body movements through interpolation factor parameters to synthesize new human body movements.

[0058] In combination with the above technical solutions and the technical problems solved, please analyze the advantages and positive effects of the technical solutions to be protected by the present invention from the following aspects:

[0059] First, in view of the technical problems existing in the above-mentioned prior art and the difficulty of solving these problems, this paper closely combines the technical solutions to be protected by the present invention and the results and data during the research and development process, and analyzes in detail and in depth how the technical solutions of the present invention solve the technical problems and some creative technical effects brought about by solving the problems. The specific description is as follows:

[0060] The present invention processes human motion capture data; a method for retrieving human motion capture data constructs a VGG network with a hash layer added before the classification layer; by simultaneously minimizing the loss function defined by the classification error and the constraints on the hash code, the learned compact binary hash code can obtain good discrimination ability and balanced bit distribution, effectively improving the retrieval accuracy and efficiency of human motion capture data; at the same time, a generation model constructed by an animation synthesis and editing method for human motion capture data provides an intuitive and convenient interactive means so that non-professional animators can also use the model to generate high-quality animations.

[0061] Second, considering the technical solution as a whole or from the perspective of the product, the technical effects and advantages of the technical solution to be protected by the present invention are described in detail as follows:

[0062] The present invention processes human motion capture data; a method for retrieving human motion capture data constructs a VGG network with a hash layer added before the classification layer; by simultaneously minimizing the loss function defined by the classification error and the constraints on the hash code, the learned compact binary hash code can obtain good discrimination ability and balanced bit distribution, effectively improving the retrieval accuracy and efficiency of human motion capture data; at the same time, a generation model constructed by an animation synthesis and editing method for human motion capture data provides an intuitive and convenient interactive means so that non-professional animators can also use the model to generate high-quality animations. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a flow chart of a method for correcting noise and errors in human motion capture data provided by an embodiment of the present invention.

[0064] Figure 2 This is a flow chart of a method for processing human motion capture data and retrieving human motion capture data provided by an embodiment of the present invention.

[0065] Figure 3 This is a flow chart of a method for performing animation synthesis and editing on human motion captured data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0067] 1. Explanatory Examples In order to enable those skilled in the art to fully understand how to implement the present invention, this section provides an illustrative example that expands upon the technical solutions of the claims.

[0068] like Figure 1 As shown, the present invention provides a method for correcting noise and errors in human motion capture data, comprising the following steps:

[0069] S101, arranging cameras in multiple directions around a moving human body to collect human body motion data through the cameras; constructing a human body geometric model through a modeling program; setting human body mass distribution parameters; and inputting the captured human body motion data;

[0070] The method for collecting human motion data by using a camera:

[0071] The human body wears a cursor that reflects a specific infrared laser, and multiple infrared laser cameras are deployed around the scene;

[0072] The laser camera will actively emit specific infrared lasers towards the human body;

[0073] When the laser hits the cursor, it reflects; the camera calculates the spatial position of each cursor by receiving these reflected lights and obtains human motion data;

[0074] The method for constructing a human body geometric model through a modeling program:

[0075] Acquire whole-body image scanning data of a human body; segment the whole-body image data of the human body to obtain human body segmentation data;

[0076] Performing three-dimensional reconstruction on the human body segmentation data to obtain a voxelized anisotropic human body geometric model; converting the voxelized anisotropic human body geometric model into a curved anisotropic human body geometric model;

[0077] Segmenting the whole-body image data to obtain the body segmentation data includes: performing structural segmentation on the whole-body image data according to the human anatomical structure, and identifying anisotropic tissues and organs with different directions in the whole-body image data;

[0078] Identifying the anisotropic tissues with different directions and the anisotropic organs with different directions as different sub-tissues and different sub-organs to obtain the segmentation data;

[0079] The converting of the voxelized anisotropic human body geometric model into a curved anisotropic human body geometric model comprises: applying a curved surface model generation algorithm to reconstruct various tissues and organs of the voxelized anisotropic human body geometric model into curved surface anisotropic human body geometric models; the curved surface model generation algorithm comprises a marching cubes algorithm;

[0080] S102, processing human motion capture data; retrieving human motion capture data; parameterizing human motion joints based on the motion data; establishing a hybrid Newton-Euler multi-rigid body dynamics equation based on Euler angles and quaternions; establishing an optimization model based on physical constraints; setting the number of iterations of the optimization calculation; solving the optimization model using a stepwise quadratic programming method; and obtaining human motion data that satisfies the physical constraints;

[0081] The method for processing human motion capture data:

[0082] Build a human body database and store the captured human body motion data into the human body database;

[0083] Calibrate human motion capture data and remove inaccurate data;

[0084] Perform statistics on human motion capture data through statistical programs; and analyze the human motion capture data;

[0085] S103, performing animation synthesis and editing on the data captured from the human body motion.

[0086] like Figure 2 As shown, the present invention provides a method for processing human motion capture data and retrieving human motion capture data as follows:

[0087] S201, constructing a human motion capture database; generating a human motion representation image for each segment of human motion capture data based on a human motion capture data sequence in the human motion capture database; constructing a deep hash feature extraction model, and extracting hash features of the human motion representation image using a pre-trained deep hash feature extraction model;

[0088] The deep hash feature extraction model is based on the deep neural network VGG16, and a hash layer activated by a sigmoid function is added before the classification layer of the deep neural network VGG16. During the training process of the deep hash feature extraction model, the following constraints are added to the loss function:

[0089] (1)

[0090] (2) J3 = tr(Out(H)(Out(H)) T )

[0091] Where Out(H) is the feature vector output by the hash layer in the deep hash feature extraction model, and e represents a one-dimensional vector with all elements equal to 1;

[0092] S202, for the human motion capture data segment to be retrieved, extract its hash feature, and calculate the distance between the hash feature of the human motion capture data to be retrieved and the hash feature of each human motion capture data segment in the human motion capture database, and select the first k data with the closest distance as the retrieval result.

[0093] The generation of the human body motion representation image provided by the present invention comprises the following steps:

[0094] Acquire human motion capture data from a human motion capture database;

[0095] Based on forward human kinematics, the joint angle information in the human motion capture data is converted into the trajectory information of each joint point of the human body;

[0096] The spatial configuration and dynamic characteristics of the joint trajectory over a period of time are represented as three channels of RGB images, which are converted from the trajectory, velocity field and self-similarity of the joint respectively.

[0097] Among the three channels of the RGB image provided by the present invention, the first channel is encoded by the trajectory of the joints, which retains the spatial information of the joints in each frame; the second channel is encoded by the velocity field, which is composed of the velocity vector of each joint and describes the human kinematic characteristics of the human body moving object; the third channel reflects the change in the distance between each joint pair, which is encoded by the self-similar matrix connecting the joints along the time dimension.

[0098] The present invention provides a method for extracting hash features of the human motion representation image using a pre-trained deep hash feature extraction model, and the specific steps are as follows:

[0099] Training the deep hash feature extraction model using human motion representation images;

[0100] The training termination condition is that the classification accuracy of the deep hash feature extraction model is greater than a preset threshold;

[0101] The features of the hash layer of the deep hash feature extraction model are extracted as hash features of the human motion representation image.

[0102] like Figure 3 As shown, the method for performing animation synthesis and editing on human motion captured data provided by the present invention is as follows:

[0103] S301: Perform simple preprocessing on the human motion capture data to annotate the prior information required for constructing the generative model; define a random process based on the annotated information, and assign a suitable kernel function to each factor based on the specific meaning of each factor to describe the mapping relationship between it and the human motion capture data, thereby modeling the human motion capture data;

[0104] S302, constructing an optimization objective function, using a numerical optimization algorithm to solve unknown parameters and build a generative model; using the obtained generative model, editing the decomposed factors to achieve synthesis and editing of human body movements.

[0105] The specific contents of the preprocessing and labeling of human motion capture data provided by the present invention are as follows:

[0106] Step (A1): Calculate the feature vector based on the human motion capture data set. The feature vector includes the root node position offset, the root node orientation, the degrees of freedom of each joint point, and the acceleration of each joint;

[0107] Step (A2) labels each human motion segment, wherein the labeling content specifically includes: person, style, content, gender, and age information.

[0108] The steps of constructing the generation model provided by the present invention are specifically as follows:

[0109] Step (B1), define a random process based on the information marked in step (1);

[0110] Step (B2): assigning a suitable kernel function to each factor in the random process defined in step (B1) according to the specific meaning of each factor. The kernel function can be any valid kernel function such as RBF function and linear function.

[0111] Step (B3), after specifying the kernel function for each factor, the covariance matrix of each factor is calculated, and then the covariance matrix of the generated model is obtained by multiplying the covariance matrices, and then the modeling of the human motion capture data is completed.

[0112] The steps of constructing the objective function to solve the unknown parameters provided by the present invention are as follows:

[0113] Step (C1), constructing a conditional probability formula and constructing the objective function using maximum a posteriori estimation (MAP);

[0114] Step (C2): using the numerical optimization algorithm SCG to solve the unknown parameters to obtain the generative model.

[0115] The specific contents of the present invention for realizing human motion synthesis and editing by using a generative model are as follows:

[0116] Content (D1), synthesize new human motion by specifying factor parameters;

[0117] Content (D2), by estimating the unknown human motion parameters, the human motion and human motion style are oriented;

[0118] Content (D3), realizes the fusion of two or more human body movements through interpolation factor parameters to synthesize new human body movements.

[0119] 2. Application Examples: In order to demonstrate the creativity and technical value of the technical solution of the present invention, this section provides application examples of the claimed technical solution on specific products or related technologies.

[0120] The present invention processes human motion capture data; a method for retrieving human motion capture data constructs a VGG network with a hash layer added before the classification layer; by simultaneously minimizing the loss function defined by the classification error and the constraints on the hash code, the learned compact binary hash code can obtain good discrimination ability and balanced bit distribution, effectively improving the retrieval accuracy and efficiency of human motion capture data; at the same time, a generation model constructed by an animation synthesis and editing method for human motion capture data provides an intuitive and convenient interactive means so that non-professional animators can also use the model to generate high-quality animations.

[0121] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0122] 3. Evidence of the effects of the embodiments: The embodiments of the present invention have achieved some positive effects during the development or use process, and indeed have great advantages over the existing technology. The following content describes them with reference to the data, charts, etc. of the experimental process.

[0123] The present invention processes human motion capture data; a method for retrieving human motion capture data constructs a VGG network with a hash layer added before the classification layer; by simultaneously minimizing the loss function defined by the classification error and the constraints on the hash code, the learned compact binary hash code can obtain good discrimination ability and balanced bit distribution, effectively improving the retrieval accuracy and efficiency of human motion capture data; at the same time, a generation model constructed by an animation synthesis and editing method for human motion capture data provides an intuitive and convenient interactive means so that non-professional animators can also use the model to generate high-quality animations.

[0124] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for correcting noise and errors in human motion capture data, characterized in that: The method for correcting noise and errors in human motion capture data comprises the following steps: Step 1: Arrange cameras in multiple directions around the human body to collect human body motion data through the cameras; construct a human body geometric model through a modeling program; set human body mass distribution parameters; and input the captured human body motion data; The method for collecting human motion data by using a camera: The human body wears a cursor that reflects a specific infrared laser, and multiple infrared laser cameras are deployed around the scene; The laser camera will actively emit specific infrared lasers towards the human body; When the laser hits the cursor, it reflects; the camera calculates the spatial position of each cursor by receiving these reflected lights and obtains human motion data; The method for constructing a human body geometric model through a modeling program: Acquire whole-body image scanning data of a human body; segment the whole-body image data of the human body to obtain human body segmentation data; Performing three-dimensional reconstruction on the human body segmentation data to obtain a voxelized anisotropic human body geometric model; converting the voxelized anisotropic human body geometric model into a curved anisotropic human body geometric model; Segmenting the whole-body image data to obtain the body segmentation data includes: performing structural segmentation on the whole-body image data according to the human anatomical structure, and identifying anisotropic tissues and organs with different directions in the whole-body image data; Identifying the anisotropic tissues with different directions and the anisotropic organs with different directions as different sub-tissues and different sub-organs to obtain the segmentation data; The converting of the voxelized anisotropic human body geometric model into a curved anisotropic human body geometric model comprises: applying a curved surface model generation algorithm to reconstruct various tissues and organs of the voxelized anisotropic human body geometric model into curved surface anisotropic human body geometric models; the curved surface model generation algorithm comprises a marching cubes algorithm; Step 2: Process the human motion capture data; retrieve the human motion capture data; parameterize the human motion joints according to the motion data; establish a hybrid Newton-Euler multi-rigid body dynamics equation based on Euler angles and quaternions; establish an optimization model based on physical constraints; set the number of iterations of the optimization calculation; solve the optimization model using a stepwise quadratic programming method; and obtain human motion data that meets the physical constraints. The method for processing human motion capture data: Build a human body database and store the captured human body motion data into the human body database; Calibrate human motion capture data and remove inaccurate data; Perform statistics on human motion capture data through statistical programs; and analyze the human motion capture data; Step three: Animation synthesis and editing of human motion capture data.

2. The method for correcting noise and errors in human motion capture data according to claim 1, wherein: The human body motion capture data is processed; the method for retrieving the human body motion capture data is as follows: (1) Constructing a human motion capture database; generating a human motion representation image for each segment of human motion capture data based on the human motion capture data sequence in the human motion capture database; constructing a deep hash feature extraction model, and extracting hash features of the human motion representation image through the pre-trained deep hash feature extraction model; The deep hash feature extraction model is based on the deep neural network VGG16, and a hash layer activated by a sigmoid function is added before the classification layer of the deep neural network VGG16. During the training process of the deep hash feature extraction model, the following constraints are added to the loss function: （1） （2） Where Out(H) is the feature vector output by the hash layer in the deep hash feature extraction model, and e represents a one-dimensional vector with all elements equal to 1; (2) For the human motion capture data segment to be retrieved, extract its hash feature, and calculate the distance between the hash feature of the human motion capture data to be retrieved and the hash feature of each segment of human motion capture data in the human motion capture database, and select the first k data with the closest distance as the retrieval result.

3. The method for correcting noise and errors in human motion capture data according to claim 2, wherein: The generation of the human body motion representation image comprises the following steps: Acquire human motion capture data from a human motion capture database; Based on forward human kinematics, the joint angle information in the human motion capture data is converted into the trajectory information of each joint point of the human body; The spatial configuration and dynamic characteristics of the joint trajectory over a period of time are represented as three channels of RGB images, which are converted from the trajectory, velocity field and self-similarity of the joint respectively.

4. The method for correcting noise and errors in human motion capture data according to claim 3, wherein: Of the three channels of the RGB image, the first channel is encoded by the joint trajectory, which retains the spatial information of the joints in each frame; the second channel is encoded by the velocity field, which consists of the velocity vector of each joint, which describes the human kinematic characteristics of the human body moving object; The third channel reflects the change in the distance between each joint pair, which is encoded by a self-similarity matrix connecting the joints along the temporal dimension.

5. The method for correcting noise and errors in human motion capture data according to claim 2, wherein: The hash features of the human motion representation image are extracted by the pre-trained deep hash feature extraction model, and the specific steps are: Training the deep hash feature extraction model using human motion representation images; The training termination condition is that the classification accuracy of the deep hash feature extraction model is greater than a preset threshold; The features of the hash layer of the deep hash feature extraction model are extracted as hash features of the human motion representation image.

6. The method for correcting noise and errors in human motion capture data according to claim 1, wherein: The method for performing animation synthesis and editing on the data captured by human motion is as follows: 1) Perform simple preprocessing on human motion capture data to annotate the prior information needed to construct the generative model. Based on the annotated information, define a random process and assign an appropriate kernel function to each factor based on its specific meaning to describe the mapping relationship between it and the human motion capture data, thereby modeling the human motion capture data. 2) Construct an optimization objective function and use a numerical optimization algorithm to solve the unknown parameters to build a generative model; using the obtained generative model, the decomposed factors are edited to achieve the synthesis and editing of human motion.

7. The method for correcting noise and errors in human motion capture data according to claim 6, wherein: The specific contents of the preprocessing and labeling of human motion capture data are as follows: Step (A1): Calculate the feature vector based on the human motion capture dataset. The feature vector includes the root node position offset, the root node orientation, the degrees of freedom of each joint point, and the acceleration of each joint. Step (A2): label each human motion segment, and the labeling content specifically includes: person, style, content, gender, and age information.

8. The method for correcting noise and errors in human motion capture data according to claim 6, wherein: The steps of constructing the generative model are as follows: Step (B1): Define a random process based on the information marked in step (1); Step (B2): assigning a suitable kernel function to each factor in the random process defined in step (B1) according to the specific meaning of each factor, wherein the kernel function is a valid kernel function, and the valid kernel functions include RBF function and linear function; Step (B3), after specifying the kernel function for each factor, the covariance matrix of each factor is calculated, and then the covariance matrix of the generative model is obtained by multiplying the covariance matrices, and then the modeling of the human motion capture data is completed.

9. The method for correcting noise and errors in human motion capture data according to claim 6, wherein: The steps to construct the objective function to solve the unknown parameters are as follows: Step (C1), construct the conditional probability formula and use the maximum a posteriori estimation MAP to construct the objective function; Step (C2): using the numerical optimization algorithm SCG to solve the unknown parameters and obtain the generative model.

10. The method for correcting noise and errors in human motion capture data according to claim 6, wherein: The specific content of using the generative model to realize human motion synthesis and editing is as follows: Content (D1), synthesize new human motion by specifying factor parameters; Content (D2): By estimating the unknown human motion parameters, the human motion and human motion style can be oriented. Content (D3), realizes the fusion of two or more human motions through interpolation factor parameters to synthesize new human motion.

Images