Animation character motion capture method and system based on machine learning

Abstract

The invention relates to the technical field of motion capture, and discloses a cartoon role motion capture method and system based on machine learning, and the method comprises the steps: collecting a spatial three-dimensional point cloud of a to-be-captured cartoon role, and purifying the point cloud to obtain a purified point cloud; the method comprises the following steps: performing human body architecture simulation on a to-be-captured animation role to obtain a simulated human body architecture, collecting inertial data, calculating a shortest cumulative distance between a point cloud frame and a data point, and constructing an initial feature matrix; setting a physical constraint of the simulated human body architecture to perform node position correction on the simulated human body architecture to obtain an optimized human body architecture; constructing an observer for optimizing the human body architecture, analyzing the initial state of the optimized human body architecture, constructing an action sequence for optimizing the human body architecture, and segmenting to obtain action subunits; and identifying the motion features to perform matching, calculating a probability value of the matched motion, and implementing motion capture of the to-be-captured cartoon character. According to the invention, the motion capture accuracy of the cartoon role can be improved.

Description

Technical Field

[0001] The present invention relates to a method and system for capturing animation character motion based on machine learning, belonging to the technical field of motion capture. Background Art

[0002] Nowadays, animation character motion capture based on machine learning is an important technical means in the field of animation production. It uses machine learning algorithms to process human motion data to drive animation characters to present realistic motion performances.

[0003] In the process of motion capture, traditional motion capture methods often rely on sensors and markers. This method has many limitations. For example, the sensor needs to be accurately fixed on a specific part of the human body, and the markers may affect the accuracy of data collection due to occlusion or environmental interference. In addition, in complex action scenes, traditional methods are difficult to accurately capture subtle movements and rapidly changing movements, resulting in a significant reduction in the fluency and naturalness of the animation character's movements. At the same time, some existing machine learning models face the problem of huge data annotation workload when processing motion capture data. Due to the diversity and complexity of motion data, accurately annotating each action sample requires a lot of manpower and time costs. Moreover, some models perform poorly in terms of generalization ability. When encountering new action types that are significantly different from the training data, they cannot be effectively captured and converted, making the animation character's action performance not rich and diverse enough. In addition, different animation characters have unique body proportions and style characteristics. When traditional motion capture technology adapts to these diverse character requirements, it requires a lot of manual adjustment and optimization, which is inefficient and has low accuracy.

[0004] Therefore, a solution is urgently needed to improve the accuracy of motion capture of animation characters. Summary of the invention

[0005] The present invention provides a method and system for capturing animation character motion based on machine learning, the main purpose of which is to improve the accuracy of security authentication while reducing the workload of path detection.

[0006] To achieve the above object, the present invention provides a method for capturing animation character motion based on machine learning, comprising: Collecting a spatial three-dimensional point cloud of a cartoon character to be captured, constructing a point cloud radius of the spatial three-dimensional point cloud, calculating statistical features of each point cloud in the spatial three-dimensional point cloud according to the point cloud radius, and performing point cloud purification on the spatial three-dimensional point cloud based on the statistical features to obtain a purified point cloud; Performing a human body structure simulation on the to-be-captured animated character to obtain a simulated human body structure, collecting inertial data of the to-be-captured animated character, constructing a distance matrix between a point cloud frame in the purified point cloud and data points in the inertial data, calculating the shortest cumulative distance between the point cloud frame and the data points using the distance matrix, aligning the point cloud frame and the data points based on the shortest cumulative distance to obtain aligned data points, and constructing an initial feature matrix of the simulated human body structure using the aligned data points; Using the initial feature matrix to set physical constraints of the simulated human body structure, wherein the physical constraints include physical motion constraints and joint length constraints, based on the physical constraints, constructing a node position optimization function of the simulated human body structure, and using the node position optimization function to correct the node positions of the simulated human body structure to obtain an optimized human body structure; querying the relative position relationship of the joint points and the change amount of the joint angles in the optimized human body structure to construct an observer of the optimized human body structure, using the observer to analyze the initial state of the optimized human body structure, constructing a state transition probability matrix of the optimized human body structure based on the initial state, constructing an action sequence of the optimized human body structure using the state transition probability matrix, and performing action segmentation on the action sequence to obtain action subunits; Identify the action features of the action subunit, construct an action label of the action feature, match the action label with a preconfigured action library to obtain a matching action, calculate a probability value of the matching action, determine a target action based on the probability value, and implement motion capture of the to-be-captured animated character based on the target action.

[0007] Optionally, the performing point cloud purification on the spatial three-dimensional point cloud based on the statistical features to obtain a purified point cloud includes: The point cloud discreteness of the spatial three-dimensional point cloud is calculated based on the statistical features using the following formula: ; in, represents the discreteness of the point cloud, represents the set of domain points of point p in the three-dimensional point cloud of space, Q represents the domain point of point p, represents the distance from point p to Q, Represents the average distance of the spatial three-dimensional point cloud in the statistical feature; Based on the point cloud discreteness, the spatial three-dimensional point cloud is purified to obtain a purified point cloud.

[0008] Optionally, constructing a distance matrix between a point cloud frame in the purified point cloud and data points in the inertial data includes: Performing time dimension conversion on the point cloud frame and the data points to obtain a converted point cloud frame and a converted data point; Determining the spatial dimensions of the converted point cloud frame and the converted data points; Constructing the spatial coordinates corresponding to the converted point cloud frame and the converted data points based on the spatial dimension to obtain point cloud frame coordinates and data point coordinates; Based on the point cloud frame coordinates and the data point coordinates, calculating the Euclidean distance between the corresponding points of the converted point cloud frame and the converted data point to obtain a distance value; The distance values ​​are stored in a pre-constructed blank matrix to obtain a distance matrix.

[0009] Optionally, the calculating the shortest cumulative distance between the point cloud frame and the data point by using the distance matrix includes: Query a point cloud sequence point in the point cloud frame and a data sequence point in the data points; Constructing the shortest cumulative distance matrix between the point cloud sequence points and the data sequence points using the distance matrix; Based on the shortest cumulative distance matrix, the shortest cumulative distance between the point cloud sequence point and the data sequence point is calculated using the following formula: ; in, represents the shortest cumulative distance, represents the shortest cumulative distance matrix, Indicates The distance between the point cloud frame and the jth data point, Indicates point cloud frame from the previous sequence point to the The distance between the previous sequence point of the data point, Indicates point cloud frame from the previous sequence point to the The distance between data points, No. point cloud frame to the The distance between the previous sequence point of a data point.

[0010] Optionally, constructing an initial feature matrix of the simulated human body architecture using the aligned data points includes: Using the aligned data points to identify joint positions corresponding to human joints; Querying the three-dimensional coordinates of the joint positions, and using the three-dimensional coordinates of the joints as joint position features of the simulated human body structure; Calculating adjacent joint distances of the joint positions, and determining limb length characteristics of the simulated human body framework based on the adjacent joint distances; Analyze the distribution status of data points in the field corresponding to the joint position; Determining joint posture angle characteristics of the simulated human body structure based on the distribution state; Organizing the data of the joint position features, the limb length features and the joint posture angle features to obtain sequence features; Data quantization is performed on the sequence features to obtain an initial feature matrix.

[0011] Optionally, constructing a node position optimization function for simulating a human body structure based on the physical constraints includes: Based on the physical constraints, construct a physical motion constraint function of the simulated human body architecture; The physical motion constraint function can be expressed as follows: ; ; in, represents the physical motion constraint function, represents the angle range constraint value, m represents the number of corresponding joint nodes in the simulated human body structure, represents the joint angle variable of joint j, express The corresponding minimum angle, express The corresponding maximum angle; Based on the physical constraints, a joint length constraint function of the simulated human body structure is constructed, wherein the joint length constraint function can be expressed by the following formula: ; ; in, represents the joint length constraint function, represents the joint length constraint value, Indicates the coordinate point sequence number of the joint in the simulated human body structure, L represents the coordinate point With coordinate points Joint lengths of connected joints; Based on the physical motion constraint function and the joint length constraint function, a node position optimization function of the simulated human body structure is determined, wherein the node position optimization function can be expressed by the following formula: ; in, represents the node position optimization function, represents the physical motion constraint function, Represents the joint length constraint function.

[0012] Optionally, the analyzing the initial state of the optimized human body architecture by using the observer includes: Using the observer, constructing the feature space of the optimized human body structure; Calculating the data point density of the feature space; Based on the data point density, construct the motion modal feature set for optimizing the human body structure Utilizing the motion modality feature set, calculating the state distribution of the optimized human body architecture; Based on the state distribution, an initial state of the optimized human body architecture is analyzed.

[0013] Optionally, constructing the state transition probability matrix of the optimized human body architecture based on the initial state includes: Setting the action sampling step length of the initial state; Based on the action sampling step, collecting a state transition data set of the initial state; Constructing a motion analysis window of the state transition data set using the motion sampling step; Identify the hidden state of each step length in the action sampling step length corresponding to the action analysis window to obtain a sequence state; Calculate the state transition frequency matrix of each state in the sequence state; Counting the total number of occurrences of each state in the state transition frequency matrix; The transition probability of each state is calculated based on the total number of times using the following formula: ; in, represents the transition probability, represents the total number of times state e appears in the state transition frequency matrix, Indicates the number of times from state e to state r; Based on the transition probabilities, a state transition probability matrix of the optimized human body architecture is constructed.

[0014] Optionally, calculating the probability value of the matching action includes: Calculating the cosine similarity between the matching action and the action features corresponding to the action subunit; Querying the occurrence frequency of the matching action; The action coordination rate of the matching action is calculated using the following formula: ; in, represents the coordination rate of the action, k represents a parameter that adjusts the steepness of the curve, represents the action coordination score of action c; According to the cosine similarity, the occurrence frequency and the action coordination rate, the probability value of the matching action is calculated using the following formula: ; ; in, represents the cosine similarity, represents the action coordination rate, represents the action coordination rate, represents the weight of cosine similarity, represents the weight of the action coordination rate, Represents the weight of the action coordination rate.

[0015] In order to solve the above problems, the present invention also provides an animation character motion capture system based on machine learning, the system comprising: A point cloud purification module is used to collect a spatial three-dimensional point cloud of a cartoon character to be captured, construct a point cloud radius of the spatial three-dimensional point cloud, calculate statistical features of each point cloud in the spatial three-dimensional point cloud according to the point cloud radius, and perform point cloud purification on the spatial three-dimensional point cloud based on the statistical features to obtain a purified point cloud; an architecture simulation module, for performing human body architecture simulation on the to-be-captured animated character to obtain a simulated human body architecture, collecting inertial data of the to-be-captured animated character, constructing a distance matrix between a point cloud frame in the purified point cloud and data points in the inertial data, calculating the shortest cumulative distance between the point cloud frame and the data points using the distance matrix, aligning the point cloud frame and the data points based on the shortest cumulative distance to obtain aligned data points, and constructing an initial feature matrix of the simulated human body architecture using the aligned data points; an architecture optimization module, configured to use the initial feature matrix to set physical constraints of the simulated human body architecture, wherein the physical constraints include physical motion constraints and joint length constraints, construct a node position optimization function of the simulated human body architecture based on the physical constraints, and use the node position optimization function to correct the node positions of the simulated human body architecture to obtain an optimized human body architecture; an action segmentation module, for querying the relative position relationship of the joint points and the change amount of the joint angles in the optimized human body structure to construct an observer of the optimized human body structure, using the observer to analyze the initial state of the optimized human body structure, constructing a state transition probability matrix of the optimized human body structure based on the initial state, constructing an action sequence of the optimized human body structure using the state transition probability matrix, and performing action segmentation on the action sequence to obtain action subunits; The motion capture module is used to identify the motion features of the motion sub-unit, construct the motion label of the motion feature, match the motion label with the preconfigured motion library to obtain the matching action, calculate the probability value of the matching action, determine the target action based on the probability value, and implement motion capture of the animation character to be captured based on the target action.

[0016] Compared with the problems described in the background technology, the present invention first collects the spatial three-dimensional point cloud of the cartoon character to be captured, constructs the point cloud radius and calculates the statistical features to purify the point cloud, thereby obtaining the accurate original data of the character in the three-dimensional space and reducing noise interference. Then, the human body structure of the character is simulated and inertial data is collected, and a distance matrix is ​​constructed to calculate the shortest cumulative distance to align the data points, and then the initial feature matrix of the simulated human body structure is constructed to achieve the fusion and synchronization of different data, and provide a high-quality data foundation for subsequent motion capture; further, the present invention uses the initial feature matrix to set the physical constraints of the simulated human body structure, constructs the node position optimization function and corrects the node position to obtain the optimized human body structure, so that the simulation conforms to the physical laws of the real human body, and enhances the authenticity and reliability of the simulation; then, the joint information of the optimized human body structure is queried to construct an observation symbol, the initial state is analyzed and a state transition probability matrix is ​​constructed to construct an action sequence, and then the action sequence is divided into action sub-units, which is convenient for in-depth analysis of the human body motion structure. Finally, the action features of the action subunit are identified and an action label is constructed, which is matched with the pre-configured action library to obtain a matching action, and its probability value is calculated to determine the target action. The action capture of the animation character is implemented based on the target action, and a variety of factors are considered to ensure that the selected action meets the actual needs, improve the realism and naturalness of the animation character's action, and improve the quality of animation works. It effectively solves many problems existing in traditional motion capture methods in terms of accuracy, data processing, and action adaptation, and provides strong support and innovation for the development of animation character motion capture technology. Therefore, the animation character motion capture method and system based on machine learning provided by the embodiment of the present invention can improve the accuracy of animation character motion capture. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of a method for capturing animation character motion based on machine learning provided by an embodiment of the present invention; Figure 2 A schematic diagram of modules for implementing the method for capturing animation character motion based on machine learning provided in one embodiment of the present invention.

[0018] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0019] 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.

[0020] The embodiment of the present application provides a method for capturing the motion of anime characters based on machine learning. The execution subject of the method for capturing the motion of anime characters based on machine learning includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for capturing the motion of anime characters based on machine learning can be executed by software or hardware installed on a terminal device or a server device. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0021] Embodiment 1: Reference Figure 1 FIG. 1 is a flow chart of a method for capturing animation character motion based on machine learning according to an embodiment of the present invention. In this embodiment, the method for capturing animation character motion based on machine learning includes: S1. Collect a spatial three-dimensional point cloud of a cartoon character to be captured, construct a point cloud radius of the spatial three-dimensional point cloud, calculate statistical features of each point cloud in the spatial three-dimensional point cloud according to the point cloud radius, and purify the spatial three-dimensional point cloud based on the statistical features to obtain a purified point cloud.

[0022] The embodiment of the present invention can obtain the original data information of the cartoon character in the three-dimensional space, such as the shape, position and other information of the character, by collecting the spatial three-dimensional point cloud of the cartoon character to be captured, so as to provide a basic basis for subsequent motion capture.

[0023] The spatial three-dimensional point cloud refers to a collection of a large number of discrete points obtained by a measuring device in a three-dimensional space, which can be collected by structured light measurement technology.

[0024] The embodiment of the present invention can clarify which points belong to the neighborhood of a certain point by constructing the point cloud radius of the spatial three-dimensional point cloud, and then divide the spatial three-dimensional point cloud into regions to facilitate identification of point cloud features.

[0025] Optionally, the point cloud radius can be obtained by first performing a preliminary analysis on the collected three-dimensional point cloud data, such as calculating a distance distribution histogram between point clouds, observing its peak and valley values, and then clustering the point clouds using a clustering algorithm (such as K-Means clustering), taking each cluster center as a reference point, and calculating the average distance from the point cloud within the cluster to the center.

[0026] The embodiment of the present invention can better understand the intrinsic characteristics of point cloud data by calculating the statistical characteristics of each point cloud in the spatial three-dimensional point cloud according to the point cloud radius, provide a basis for judging whether the point cloud is a noise point or an outlier, thereby realizing preliminary analysis and screening of the original point cloud data.

[0027] The statistical features refer to the average distance and distance standard deviation of the spatial three-dimensional point cloud.

[0028] Optionally, the statistical features of each point cloud in the spatial three-dimensional point cloud calculated according to the point cloud radius can be obtained by determining the neighborhood point set of each point with the point cloud radius, calculating the distance between the point and its neighborhood points and summing them, and then dividing the distance by the number of neighborhood points to obtain the average distance. For the distance standard deviation, the sum of the squares of the difference between the distance between the point and the neighborhood points and the average distance is first calculated, and the sum is divided by the number of neighborhood points and then the square root is taken to obtain the standard deviation.

[0029] The embodiment of the present invention purifies the spatial three-dimensional point cloud based on the statistical features to obtain a purified point cloud, which can reduce the interference of noise and outliers on subsequent processing and improve the accuracy and reliability of the entire motion capture process.

[0030] As an embodiment of the present invention, the point cloud purification of the spatial three-dimensional point cloud based on the statistical features to obtain the purified point cloud includes: calculating the point cloud discreteness of the spatial three-dimensional point cloud using the following formula based on the statistical features: ; in, represents the discreteness of the point cloud, represents the set of domain points of point p in the three-dimensional point cloud of space, Q represents the domain point of point p, represents the distance from point p to Q, Represents the average distance of the spatial three-dimensional point cloud in the statistical feature; Based on the point cloud discreteness, the spatial three-dimensional point cloud is purified to obtain a purified point cloud.

[0031] Optionally, the process of purifying the spatial three-dimensional point cloud based on the point cloud discreteness to obtain the purified point cloud is as follows: first determine the discreteness threshold according to the shape characteristics of the cartoon character, data quality requirements and subsequent application scenarios. For example, if a cartoon character with a relatively regular appearance is to be motion captured, a relatively low discreteness threshold can be set for the point cloud of the main body parts (such as the trunk and limbs), because the point cloud distribution of these parts is relatively uniform; and for complex structures such as hair, the threshold can be appropriately increased, and all points in the spatial three-dimensional point cloud are traversed. For each point, the calculated discreteness is compared with the set threshold. If the discreteness of the point exceeds the threshold, the point is determined to be a noise point or an outlier, and it is removed from the point cloud; if the discreteness is within the threshold range, the point is retained. After such a screening process, the remaining points constitute the purified point cloud.

[0032] S2. Perform human body structure simulation on the animated character to be captured to obtain a simulated human body structure, collect inertial data of the animated character to be captured, construct a distance matrix between the point cloud frame in the purified point cloud and the data points in the inertial data, calculate the shortest cumulative distance between the point cloud frame and the data points using the distance matrix, align the point cloud frame and the data points based on the shortest cumulative distance to obtain aligned data points, and construct an initial feature matrix of the simulated human body structure using the aligned data points.

[0033] The embodiment of the present invention simulates the human body structure of the animated character to be captured to obtain a simulated human body structure, which can present the character point cloud data in a more logical and structured human body framework form, making it easier to correspond to the real human body movement logic, so that the motion capture results can be more in line with the actual human body movement laws.

[0034] The simulated human body framework refers to a virtual model used to represent the structure and movement relationship of the human body. It is based on the connection mode of human bones and joints, and constructs a structure similar to the real human skeleton framework in a computer environment.

[0035] Optionally, the simulation of human body structure can be achieved by utilizing a pre-built universal human body model library (these models have standard human bone structure and joint connection methods), matching and adjusting the approximate dimensions of the animation character (such as height, limb length ratio, etc.) with the model, and adapting the universal model to the animation character to achieve human body structure simulation. The pre-built universal human body model library can be configured using a generative adversarial network (GAN) or a variational autoencoder (VAE) in deep learning.

[0036] The embodiment of the present invention can obtain the dynamic changes of the character in space, such as the speed and direction of limb swinging, by collecting the inertial data of the animated character to be captured, so as to more comprehensively depict the movement process of the character, thereby improving the accuracy and completeness of motion capture.

[0037] The inertial data refers to information about the motion state of an object, mainly including acceleration data and angular velocity data, which can be collected by an inertial measurement unit (IMU).

[0038] Furthermore, the embodiment of the present invention can establish a quantitative correspondence between two different types of data in terms of spatial position relationship by constructing a distance matrix between the point cloud frame in the purified point cloud and the data points in the inertial data, so as to find the relationship between the two and realize collaborative processing of data.

[0039] As an embodiment of the present invention, the construction of a distance matrix between the point cloud frame in the purified point cloud and the data points in the inertial data includes: performing time dimension conversion on the point cloud frame and the data points to obtain a converted point cloud frame and converted data points, determining the spatial dimensions of the converted point cloud frame and the converted data points, constructing the spatial coordinates corresponding to the converted point cloud frame and the converted data points based on the spatial dimensions to obtain point cloud frame coordinates and data point coordinates, and calculating the Euclidean distance between corresponding points of the converted point cloud frame and the converted data points based on the point cloud frame coordinates and the data point coordinates to obtain a distance value, and storing the distance value in a pre-constructed blank matrix to obtain a distance matrix.

[0040] Optionally, the point cloud frame and the data point are converted in time dimension to obtain the converted point cloud frame and the converted data point. The point cloud frame and the data point can be represented uniformly in time scale by interpolation, resampling, etc. based on the timestamp and other information of the data, so that they are implemented under the same time dimension standard. The spatial dimension can be determined by analyzing the spatial range involved in the data to determine whether it is a two-dimensional or three-dimensional space situation. Usually, such data describing the shape of an object is determined as a three-dimensional space dimension. The spatial coordinates can be obtained by extracting or converting the specific coordinate values ​​of each point cloud frame and data point in the space from the corresponding data according to the determined spatial dimension (such as three-dimensional corresponds to X, Y, and Z axes).

[0041] The embodiment of the present invention can determine the most reasonable matching method by using the distance matrix to calculate the shortest cumulative distance between the point cloud frame and the data points, so that the point cloud frame and the inertial data points can be subsequently aligned in the most suitable order and position, reducing the motion capture error caused by data asynchrony and mismatch, and ensuring the continuity and accuracy of motion capture.

[0042] It should be explained that the shortest cumulative distance refers to an optimal cumulative distance metric calculated by dynamic programming when aligning two time series, and this distance represents a "cost" of matching the two sequences in the time dimension. The cumulative distance corresponding to the path with the minimum cost is the shortest cumulative distance.

[0043] As an embodiment of the present invention, the use of the distance matrix to calculate the shortest cumulative distance between the point cloud frame and the data point includes: querying a point cloud sequence point in the point cloud frame and a data sequence point in the data point, using the distance matrix to construct a shortest cumulative distance matrix between the point cloud sequence point and the data sequence point, and based on the shortest cumulative distance matrix, using the following formula to calculate the shortest cumulative distance between the point cloud sequence point and the data sequence point: ; in, represents the shortest cumulative distance, represents the shortest cumulative distance matrix, Indicates The distance between the point cloud frame and the jth data point, Indicates point cloud frame from the previous sequence point to the The distance between the previous sequence point of the data point, Indicates point cloud frame from the previous sequence point to the The distance between data points, No. point cloud frame to the The distance between the previous sequence point of a data point.

[0044] Optionally, the point cloud sequence point can determine a specific point cloud sequence point from the point cloud frame data set, and the point cloud sequence point can be a representative point or a starting point selected in a certain order (such as time order or spatial position order), and the data sequence point can select a corresponding starting data sequence point from the data point data set. The shortest cumulative distance matrix can be obtained by gradually calculating the shortest cumulative distance of each position from the starting point based on the selected point cloud sequence point and data sequence point, using the distance matrix according to the idea of ​​dynamic programming.

[0045] The embodiment of the present invention performs data point alignment on the point cloud frame and the data points based on the shortest cumulative distance to obtain aligned data points, which can achieve fusion and synchronization of data from different sources, and provide a high-quality data foundation for more accurately constructing an initial feature matrix of the human body structure.

[0046] Optionally, the process of aligning the point cloud frame and the data points based on the shortest cumulative distance to obtain the aligned data points is as follows: starting from the lower right corner of the calculated shortest cumulative distance matrix, backtracking according to the path information recorded when constructing the matrix (i.e., whether each element comes from the left, the top, or the upper left), to obtain the shortest path from the starting point to the end point. The point pairs on this path represent the optimal alignment relationship between the point cloud frame and the data points in the time or space sequence, and the points in the point cloud frame and the points in the data points are matched one by one according to the point pair relationship on the shortest path.

[0047] The embodiment of the present invention can help users analyze the state of the human body structure, identify joints, and provide standardized data input for operations such as motion classification by using the aligned data points to construct the initial feature matrix of the simulated human body structure. It is an important link in the entire motion capture process and helps to promote the smooth implementation of subsequent motion capture and analysis work.

[0048] As an embodiment of the present invention, the initial feature matrix of the simulated human body structure is constructed using the aligned data points, including: using the aligned data points to identify joint positions corresponding to human joints, querying the three-dimensional joint coordinates of the joint positions, using the three-dimensional joint coordinates as joint position features of the simulated human body structure, calculating adjacent joint distances of the joint positions, determining limb length features of the simulated human body structure based on the adjacent joint distances, analyzing the distribution status of field data points corresponding to the joint positions, determining joint posture angle features of the simulated human body structure based on the distribution status, organizing the joint position features, the limb length features and the joint posture angle features to obtain sequence features, and quantizing the sequence features to obtain an initial feature matrix.

[0049] Optionally, the joint positions corresponding to the human joints can be identified by using the aligned data points according to the relative position relationship and movement pattern of the human joints in three-dimensional space, and a joint recognition model can be established by learning and analyzing a large amount of human motion data, and then the aligned data points are input into the model. The model determines the point cloud area corresponding to the human joint according to the spatial distribution, density change and movement trend of the data points, thereby identifying the joint position. The calculation of the adjacent joint distance of the joint position can be performed after determining the coordinates of each joint position, and the distance between them is calculated using the distance formula between two points in space for adjacent joints. The distribution of the data points corresponding to the joint position can be analyzed by taking the joint position as the center, selecting the neighborhood data points within a certain range, and observing the distribution of these data points in space. The distribution state is analyzed by calculating the vector angle between the data point and the joint position, the point cloud density change and other indicators. For example, when the knee joint is bent, the distribution of the data points before and after the joint will show obvious angle changes and density differences, and the posture information such as the bending angle of the joint is determined based on these characteristics. The data organization of the joint position features, the limb length features and the joint posture angle features can be implemented in a logical order of the human body structure, usually starting from the head joints, along the spine, upper limbs, lower limbs, etc., and arranging and combining the position features, limb length features and joint posture angle features of each joint in sequence.

[0050] S3. Using the initial feature matrix to set the physical constraints of the simulated human body structure, wherein the physical constraints include physical motion constraints and joint length constraints. Based on the physical constraints, a node position optimization function of the simulated human body structure is constructed. The node position optimization function is used to correct the node positions of the simulated human body structure to obtain an optimized human body structure.

[0051] The embodiment of the present invention can introduce physical constraints by using the initial feature matrix to set the physical constraints of the simulated human body structure, so that the movement and morphological changes of the simulated human body structure follow the physiological laws of the real human body, avoid unreasonable movements or body deformations, improve the authenticity and reliability of the simulation, and lay the foundation for subsequent accurate motion capture and animation production.

[0052] Among them, the physical constraints include physical motion constraints and joint length constraints. It should be further explained that the physical constraints define the physical limitations of the range and mode of motion that the human joints can perform in all directions, and the joint length constraints refer to setting the bone length limit of each joint connection based on the fixed length characteristics of the human skeletal structure.

[0053] Optionally, physical constraints can extract joint posture angles from the initial feature matrix to determine joint angle range constraints, and define limb movement direction constraints based on limb length and joint position; then extract limb length data to determine standard joint length values, and use the distance formula between two points to establish an equation to set long physical constraints.

[0054] Furthermore, the embodiment of the present invention can provide a quantitative means to control the change of the node position of the simulated human body structure by constructing the node position optimization function of the simulated human body structure based on the physical constraints, so that the node position can be automatically adjusted according to these functional relationships in subsequent calculations and simulations to meet the physical constraint requirements. This helps to achieve automated and accurate human body structure simulation and optimization, and improve the efficiency and accuracy of the entire motion capture and simulation system.

[0055] As an embodiment of the present invention, constructing the node position optimization function of the simulated human body structure based on the physical constraints includes: constructing a physical motion constraint function of the simulated human body structure based on the physical constraints, wherein the physical motion constraint function can be expressed by the following expression: ; ; in, represents the physical motion constraint function, represents the angle range constraint value, m represents the number of corresponding joint nodes in the simulated human body structure, represents the joint angle variable of joint j, express The corresponding minimum angle, express The corresponding maximum angle; Based on the physical constraints, a joint length constraint function of the simulated human body structure is constructed, wherein the joint length constraint function can be expressed by the following formula: ; ; in, represents the joint length constraint function, represents the joint length constraint value, Indicates the coordinate point sequence number of the joint in the simulated human body structure, L represents the coordinate point With coordinate points Joint lengths of connected joints; Based on the physical motion constraint function and the joint length constraint function, a node position optimization function of the simulated human body structure is determined, wherein the node position optimization function can be expressed by the following formula: ; in, represents the node position optimization function, represents the physical motion constraint function, Represents the joint length constraint function.

[0056] The embodiment of the present invention uses the node position optimization function to correct the node positions of the simulated human body structure to obtain an optimized human body structure, which can make the simulated human body structure fully comply with the set physical constraints and eliminate the problem of not complying with the physical characteristics of the real human body due to data errors, calculation deviations or incomplete initial conditions.

[0057] Optionally, the process of using the node position optimization function to correct the node positions of the simulated human body structure to obtain the optimized human body structure is as follows: first, the initial values ​​of the node positions and joint angles of the simulated human body structure are obtained, and the joint angle range, length and other parameters in the node position optimization function are determined. Then, the gradient descent method is used to calculate the gradient of the optimization function with respect to the joint angle variables and the node coordinate variables. According to the set step size, these variables are iteratively updated in the opposite direction of the gradient. After each iteration, whether the function value change or the gradient norm is less than a threshold is determined to determine whether it has converged, until convergence results in an optimized human body structure that meets physical constraints.

[0058] S4. Query the relative position relationship of the joint points and the change of the joint angles in the optimized human body structure to construct an observer of the optimized human body structure, use the observer to analyze the initial state of the optimized human body structure, construct a state transition probability matrix of the optimized human body structure based on the initial state, use the state transition probability matrix to construct an action sequence of the optimized human body structure, and segment the action sequence to obtain action sub-units.

[0059] The embodiment of the present invention can query the relative position relationship of the joint points and the change of the joint angles in the optimized human body structure to construct an observation symbol of the optimized human body structure, which can integrate and abstract information such as the relative position relationship of the joint points and the change of the joint angles to form a symbol or variable set that can represent the current state of the optimized human body structure, thereby conveniently describing the state of the human body structure.

[0060] Among them, the observation symbol refers to a symbol used to describe the state of the optimized human body structure. It integrates key information such as the relative position relationship of the joint points and the change in joint angles queried from the optimized human body structure. It can be regarded as a "state label" that can concisely and effectively summarize the state characteristics of the human body structure at a certain moment.

[0061] Optionally, the querying of the relative position relationship and joint angle change of the joint points in the optimized human body structure to construct the observation symbol of the optimized human body structure can obtain the joint point coordinate data through a motion capture device or sensor, and calculate the coordinate difference of adjacent joint points to obtain the relative position relationship; use an angle measurement tool or a vector calculation method based on coordinate data to measure the joint angle change; finally, integrate and quantify these data and use them as components of the observation symbol to construct the observation symbol.

[0062] Furthermore, the embodiment of the present invention can better understand the movement starting point of the human body structure by analyzing the initial state of the optimized human body structure using the observer, and then analyze its subsequent changes and developments.

[0063] As an embodiment of the present invention, the use of the observer to analyze the initial state of the optimized human body structure includes: using the observer to construct a feature space of the optimized human body structure, calculating the data point density of the feature space, constructing a motion modal feature set of the optimized human body structure based on the data point density, using the motion modal feature set to calculate the state distribution of the optimized human body structure, and analyzing the initial state of the optimized human body structure based on the state distribution.

[0064] Optionally, the feature space can use observation symbols such as the relative position relationship of joint points and the change in joint angles as dimensions to combine into a joint point feature vector that describes the state of the human body structure, and the feature vector can be used to construct the feature space. The data point density can use a density estimation algorithm (such as kernel density estimation) to determine the density around each data point in the feature space. The motion mode feature set can be obtained by grouping the feature vectors of areas with similar density according to the data point density to form a feature set representing different motion modes. The calculation of the state distribution of the optimized human body structure can be obtained by counting the proportion of each motion mode feature set in the data to obtain the probability distribution of the optimized human body structure in different motion modes (states). The initial state can be selected according to the state distribution, and the state corresponding to the motion mode with the highest probability is selected as the initial state of the optimized human body structure.

[0065] The embodiment of the present invention can quantitatively describe the possibility of the human body structure switching between different actions and postures by constructing the state transition probability matrix of the optimized human body structure based on the initial state, providing an important basis for predicting subsequent actions and postures of the human body.

[0066] As an embodiment of the present invention, the state transition probability matrix of the optimized human body structure is constructed based on the initial state, including: setting the action sampling step of the initial state, based on the action sampling step, collecting the state transition data set of the initial state, using the action sampling step to construct the action analysis window of the state transition data set, identifying the hidden state of each step in the action sampling step corresponding to the action analysis window, obtaining the sequence state, calculating the state transition frequency matrix of each state in the sequence state, counting the total number of occurrences of each state in the state transition frequency matrix, and calculating the transition probability of each state based on the total number of occurrences using the following formula: ; in, represents the transition probability, represents the total number of times state e appears in the state transition frequency matrix, Indicates the number of times from state e to state r; Based on the transition probabilities, a state transition probability matrix of the optimized human body architecture is constructed.

[0067] Optionally, the action sampling step may determine a suitable time interval as the sampling step according to the characteristics of the action data and the analysis requirements, such as sampling once every 0.2 seconds for subsequent data collection and analysis. The state transition data set may be composed of relevant action data in the initial state extracted from the data containing the optimized human body structure action according to the set action sampling step. The action analysis window is based on the action sampling step to determine the window length (such as a window containing 5 sampling steps), and the data set is divided accordingly to form action analysis windows. The sequence state may determine the hidden state of the action corresponding to each sampling step in each action analysis window using a hidden Markov model, and combine these hidden states arranged in sequence into a sequence state. The state transition frequency matrix may count the obtained sequence states, record the number of transitions from one state to another, and sort them out. The state transition probability matrix may be completed by filling the calculated state transition probabilities into a preset blank matrix according to the corresponding positions.

[0068] The embodiment of the present invention can intuitively display the movement process and behavior pattern of the human body structure over a period of time by constructing the action sequence of optimizing the human body structure using the state transition probability matrix, which helps to grasp and understand the overall complex movements of the human body.

[0069] Optionally, the action sequence for optimizing the human body structure by using the state transition probability matrix can be constructed with the initial state as the starting point, and the next state is randomly selected according to the state transition probability of each state in the state transition probability matrix, and this process is repeated continuously. Then, the selected states are connected in sequence, each state corresponds to the motion information of a specific joint point, and the action sequence for optimizing the human body structure is constructed by combining them.

[0070] The embodiment of the present invention divides the action sequence into action sub-units to obtain action sub-units. The continuous action sequence can be divided into a number of relatively independent action sub-units with specific semantics and functions according to certain standards and methods, which is convenient for more detailed analysis of the structure and composition of human body movements and for classification, identification and research of different types of actions.

[0071] Optionally, the action sequence is segmented to obtain action sub-units according to the change of state in the action sequence. When the state changes significantly, such as from "walking" to "jumping", the state switching point is determined as the segmentation point, and the action sequence is truncated, thereby obtaining action sub-units with relatively independent semantics, such as walking sub-units, jumping sub-units, etc.

[0072] S5. Identify the action features of the action subunit, construct an action label of the action feature, match the action label with a preconfigured action library to obtain a matching action, calculate a probability value of the matching action, determine a target action based on the probability value, and implement motion capture of the to-be-captured animated character based on the target action.

[0073] The embodiment of the present invention can visualize the action sub-unit from a relatively abstract state sequence into a describable and quantifiable feature set by identifying the action features of the action sub-unit.

[0074] Optionally, the action features can be obtained by using a deep learning model to extract key information such as the motion trajectory of the action sub-unit joints and speed changes.

[0075] Furthermore, the embodiment of the present invention can assign a label with clear semantics to each action subunit by constructing the action label of the action feature, so as to facilitate subsequent data management, retrieval and matching operations.

[0076] Optionally, the action label can be automatically constructed using a label script generated by Java, such as 01 for "walking" and 001 for "jumping", which can be set in combination with actual applications.

[0077] The embodiment of the present invention performs action matching between the action tag and the preconfigured action library to obtain the matching action, which can achieve rapid classification and accurate positioning of the currently captured action sub-unit.

[0078] Optionally, the matching action can be quickly filtered out in the action library through the tag index to select actions with the same tag as the matching action.

[0079] The embodiment of the present invention can quantitatively compare and evaluate multiple matching actions by calculating the probability value of the matching action, measure the suitability of each action for becoming the final target action in a scientific way, avoid the one-sidedness and uncertainty caused by selecting actions based solely on subjective judgment or simple rules, and provide an objective basis for accurately determining the actions ultimately used for the animation character.

[0080] As an embodiment of the present invention, the calculating the probability value of the matching action includes: calculating the cosine similarity between the matching action and the action feature corresponding to the action subunit, querying the occurrence frequency of the matching action, and calculating the action coordination rate of the matching action using the following formula: ; in, represents the coordination rate of the action, k represents a parameter that adjusts the steepness of the curve, represents the action coordination score of action c; According to the cosine similarity, the occurrence frequency and the action coordination rate, the probability value of the matching action is calculated using the following formula: ; ; in, represents the cosine similarity, represents the action coordination rate, represents the action coordination rate, represents the weight of cosine similarity, represents the weight of the action coordination rate, Represents the weight of the action coordination rate.

[0081] It should be explained that the action coordination score is obtained by learning the coordination pattern between action sequences through a machine learning model (such as a recurrent neural network), taking the entire action sequence (including the current action sub-unit and its previous and subsequent actions) and the matching action as input.

[0082] The embodiment of the present invention can ensure that the final selected action is the most consistent with the actual capture situation and animation production requirements after comprehensive consideration of multiple factors by determining the target action based on the probability value, so that the action performance of the animation character is consistent with the logic of human movement and can be coordinated with the overall action style, scene, etc., thereby enhancing the realism and naturalness of the animation character's action and improving the quality of the animation work.

[0083] Optionally, the target action may compare probability values, and the matching action corresponding to the maximum probability value is selected as the target action.

[0084] The embodiment of the present invention can accurately map the determined target action to the animated character to be captured by implementing motion capture of the animated character to be captured based on the target action, allowing the animated character to perform corresponding joint movements, posture changes, etc. according to the action, thereby realizing the transfer of motion data from actual capture to the virtual character, and completing the final implementation link of motion capture.

[0085] Optionally, the motion capture of the to-be-captured animated character based on the target motion can deeply analyze the skeletal structure of the to-be-captured animated character, and accurately correspond the joint points in the target motion data to the skeletal joints of the character. Then coordinate conversion and scaling are performed to adapt the motion data to the character coordinate system and proportion. Then the motion data is interpolated and smoothed to enhance the continuity and smoothness of the motion. Finally, the processed data is assigned to the skeletal joint attributes of the character frame by frame through the programming interface, driving the character to smoothly present the animation effect according to the target motion in the virtual scene.

[0086] Embodiment 2: like Figure 2The figure shows a functional module diagram of a cartoon character motion capture system based on machine learning according to the present invention.

[0087] The animation character motion capture system 200 based on machine learning described in the present invention can be installed in an electronic device. According to the functions implemented, the animation character motion capture system based on machine learning can include a point cloud purification module 201, an architecture simulation module 202, an architecture optimization module 203, an action segmentation module 204 and a motion capture module 205. The module described in the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0088] In the embodiment of the present invention, the functions of each module / unit are as follows: A point cloud purification module is used to collect a spatial three-dimensional point cloud of a cartoon character to be captured, construct a point cloud radius of the spatial three-dimensional point cloud, calculate statistical features of each point cloud in the spatial three-dimensional point cloud according to the point cloud radius, and perform point cloud purification on the spatial three-dimensional point cloud based on the statistical features to obtain a purified point cloud; an architecture simulation module, for performing human body architecture simulation on the to-be-captured animated character to obtain a simulated human body architecture, collecting inertial data of the to-be-captured animated character, constructing a distance matrix between a point cloud frame in the purified point cloud and data points in the inertial data, calculating the shortest cumulative distance between the point cloud frame and the data points using the distance matrix, aligning the point cloud frame and the data points based on the shortest cumulative distance to obtain aligned data points, and constructing an initial feature matrix of the simulated human body architecture using the aligned data points; an architecture optimization module, configured to use the initial feature matrix to set physical constraints of the simulated human body architecture, wherein the physical constraints include physical motion constraints and joint length constraints, construct a node position optimization function of the simulated human body architecture based on the physical constraints, and use the node position optimization function to correct the node positions of the simulated human body architecture to obtain an optimized human body architecture; an action segmentation module, for querying the relative position relationship of the joint points and the change amount of the joint angles in the optimized human body structure to construct an observer of the optimized human body structure, using the observer to analyze the initial state of the optimized human body structure, constructing a state transition probability matrix of the optimized human body structure based on the initial state, constructing an action sequence of the optimized human body structure using the state transition probability matrix, and performing action segmentation on the action sequence to obtain action subunits; The motion capture module is used to identify the motion features of the motion sub-unit, construct the motion label of the motion feature, match the motion label with the preconfigured motion library to obtain the matching action, calculate the probability value of the matching action, determine the target action based on the probability value, and implement motion capture of the animation character to be captured based on the target action.

[0089] In detail, each module in the animation character motion capture system 200 based on machine learning in the embodiment of the present invention is used in the same manner as described above. Figure 1 The same technical means are used as the animation character motion capture method based on machine learning described in , and can produce the same technical effects, so I will not go into details here.

[0090] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for capturing animation character motion based on machine learning, characterized in that: The method comprises: Collecting a spatial three-dimensional point cloud of a cartoon character to be captured, constructing a point cloud radius of the spatial three-dimensional point cloud, calculating statistical features of each point cloud in the spatial three-dimensional point cloud according to the point cloud radius, and performing point cloud purification on the spatial three-dimensional point cloud based on the statistical features to obtain a purified point cloud; Performing a human body structure simulation on the to-be-captured animated character to obtain a simulated human body structure, collecting inertial data of the to-be-captured animated character, constructing a distance matrix between a point cloud frame in the purified point cloud and data points in the inertial data, calculating the shortest cumulative distance between the point cloud frame and the data points using the distance matrix, aligning the point cloud frame and the data points based on the shortest cumulative distance to obtain aligned data points, and constructing an initial feature matrix of the simulated human body structure using the aligned data points; Using the initial feature matrix to set physical constraints of the simulated human body structure, wherein the physical constraints include physical motion constraints and joint length constraints, based on the physical constraints, constructing a node position optimization function of the simulated human body structure, and using the node position optimization function to correct the node positions of the simulated human body structure to obtain an optimized human body structure; querying the relative position relationship of the joint points and the change amount of the joint angles in the optimized human body structure to construct an observer of the optimized human body structure, using the observer to analyze the initial state of the optimized human body structure, constructing a state transition probability matrix of the optimized human body structure based on the initial state, constructing an action sequence of the optimized human body structure using the state transition probability matrix, and performing action segmentation on the action sequence to obtain action subunits; Identify the action features of the action subunit, construct an action label of the action feature, match the action label with a preconfigured action library to obtain a matching action, calculate a probability value of the matching action, determine a target action based on the probability value, and implement motion capture of the to-be-captured animated character based on the target action.

2. The method for capturing animation character motion based on machine learning as claimed in claim 1, characterized in that: The step of purifying the spatial three-dimensional point cloud based on the statistical features to obtain a purified point cloud includes: The point cloud discreteness of the spatial three-dimensional point cloud is calculated based on the statistical features using the following formula: ； in, represents the discreteness of the point cloud, represents the set of domain points of point p in the three-dimensional point cloud of space, Q represents the domain point of point p, represents the distance from point p to Q, Represents the average distance of the spatial three-dimensional point cloud in the statistical feature; Based on the point cloud discreteness, the spatial three-dimensional point cloud is purified to obtain a purified point cloud.

3. The method for capturing animation character motion based on machine learning as claimed in claim 1, characterized in that: The step of constructing a distance matrix between a point cloud frame in the purified point cloud and data points in the inertial data includes: Performing time dimension conversion on the point cloud frame and the data points to obtain a converted point cloud frame and a converted data point; Determining the spatial dimensions of the converted point cloud frame and the converted data points; Constructing the spatial coordinates corresponding to the converted point cloud frame and the converted data points based on the spatial dimension to obtain point cloud frame coordinates and data point coordinates; Based on the point cloud frame coordinates and the data point coordinates, calculating the Euclidean distance between the corresponding points of the converted point cloud frame and the converted data point to obtain a distance value; The distance values ​​are stored in a pre-constructed blank matrix to obtain a distance matrix.

4. The method for capturing animation character motion based on machine learning as claimed in claim 1, characterized in that: The calculating the shortest cumulative distance between the point cloud frame and the data point by using the distance matrix comprises: Query a point cloud sequence point in the point cloud frame and a data sequence point in the data points; Constructing the shortest cumulative distance matrix between the point cloud sequence points and the data sequence points using the distance matrix; Based on the shortest cumulative distance matrix, the shortest cumulative distance between the point cloud sequence point and the data sequence point is calculated using the following formula: ； in, represents the shortest cumulative distance, represents the shortest cumulative distance matrix, Indicates The distance between the point cloud frame and the jth data point, Indicates point cloud frame from the previous sequence point to the The distance between the previous sequence point of the data point, Indicates point cloud frame from the previous sequence point to the The distance between data points, No. point cloud frame to the The distance between the previous sequence point of a data point.

5. The method for capturing animation character motion based on machine learning as claimed in claim 1, characterized in that: The initial feature matrix of the simulated human body architecture is constructed using the aligned data points, including: Using the aligned data points to identify joint positions corresponding to human joints; Querying the three-dimensional coordinates of the joint positions, and using the three-dimensional coordinates of the joints as joint position features of the simulated human body structure; Calculating adjacent joint distances of the joint positions, and determining limb length characteristics of the simulated human body framework based on the adjacent joint distances; Analyze the distribution status of data points in the field corresponding to the joint position; Determining joint posture angle characteristics of the simulated human body structure based on the distribution state; Organizing the data of the joint position features, the limb length features and the joint posture angle features to obtain sequence features; Data quantization is performed on the sequence features to obtain an initial feature matrix.

6. The method for capturing animation character motion based on machine learning as claimed in claim 1, characterized in that: The step of constructing a node position optimization function for simulating a human body structure based on the physical constraints includes: Based on the physical constraints, construct a physical motion constraint function of the simulated human body architecture; The physical motion constraint function can be expressed as follows: ； ； in, represents the physical motion constraint function, represents the angle range constraint value, m represents the number of corresponding joint nodes in the simulated human body structure, represents the joint angle variable of joint j, express The corresponding minimum angle, express The corresponding maximum angle; Based on the physical constraints, a joint length constraint function of the simulated human body structure is constructed, wherein the joint length constraint function can be expressed by the following formula: ； ； in, represents the joint length constraint function, represents the joint length constraint value, Indicates the coordinate point sequence number of the joint in the simulated human body structure, L represents the coordinate point With coordinate points Joint lengths of connected joints; Based on the physical motion constraint function and the joint length constraint function, a node position optimization function of the simulated human body structure is determined, wherein the node position optimization function can be expressed by the following formula: ； in, represents the node position optimization function, represents the physical motion constraint function, Represents the joint length constraint function.

7. The method for capturing animation character motion based on machine learning as claimed in claim 1, characterized in that: The analyzing the initial state of the optimized human body structure by using the observer includes: Using the observer, constructing the feature space of the optimized human body structure; Calculating the data point density of the feature space; Based on the data point density, construct the motion modal feature set for optimizing the human body structure Utilizing the motion modality feature set, calculating the state distribution of the optimized human body architecture; Based on the state distribution, an initial state of the optimized human body architecture is analyzed.

8. The method for capturing animation character motion based on machine learning as claimed in claim 1, characterized in that: The step of constructing the state transition probability matrix of the optimized human body architecture based on the initial state includes: Setting the action sampling step length of the initial state; Based on the action sampling step, collecting a state transition data set of the initial state; Constructing a motion analysis window of the state transition data set using the motion sampling step; Identify the hidden state of each step length in the action sampling step length corresponding to the action analysis window to obtain a sequence state; Calculate the state transition frequency matrix of each state in the sequence state; Counting the total number of occurrences of each state in the state transition frequency matrix; The transition probability of each state is calculated based on the total number of times using the following formula: ； in, represents the transition probability, represents the total number of times state e appears in the state transition frequency matrix, Indicates the number of times from state e to state r; Based on the transition probabilities, a state transition probability matrix of the optimized human body architecture is constructed.

9. The method for capturing animation character motion based on machine learning as claimed in claim 1, characterized in that: The calculating the probability value of the matching action includes: Calculating the cosine similarity between the matching action and the action features corresponding to the action subunit; Querying the occurrence frequency of the matching action; The action coordination rate of the matching action is calculated using the following formula: ； in, represents the coordination rate of the action, k represents a parameter that adjusts the steepness of the curve, represents the action coordination score of action c; According to the cosine similarity, the occurrence frequency and the action coordination rate, the probability value of the matching action is calculated using the following formula: ； ； in, represents the cosine similarity, represents the action coordination rate, represents the action coordination rate, represents the weight of cosine similarity, represents the weight of the action coordination rate, Represents the weight of the action coordination rate.

10. A machine learning-based animation character motion capture system, characterized in that: The system comprises: A point cloud purification module is used to collect a spatial three-dimensional point cloud of a cartoon character to be captured, construct a point cloud radius of the spatial three-dimensional point cloud, calculate statistical features of each point cloud in the spatial three-dimensional point cloud according to the point cloud radius, and perform point cloud purification on the spatial three-dimensional point cloud based on the statistical features to obtain a purified point cloud; an architecture simulation module, for performing human body architecture simulation on the to-be-captured animated character to obtain a simulated human body architecture, collecting inertial data of the to-be-captured animated character, constructing a distance matrix between a point cloud frame in the purified point cloud and data points in the inertial data, calculating the shortest cumulative distance between the point cloud frame and the data points using the distance matrix, aligning the point cloud frame and the data points based on the shortest cumulative distance to obtain aligned data points, and constructing an initial feature matrix of the simulated human body architecture using the aligned data points; an architecture optimization module, configured to use the initial feature matrix to set physical constraints of the simulated human body architecture, wherein the physical constraints include physical motion constraints and joint length constraints, construct a node position optimization function of the simulated human body architecture based on the physical constraints, and use the node position optimization function to correct the node positions of the simulated human body architecture to obtain an optimized human body architecture; an action segmentation module, for querying the relative position relationship of the joint points and the change amount of the joint angles in the optimized human body structure to construct an observer of the optimized human body structure, using the observer to analyze the initial state of the optimized human body structure, constructing a state transition probability matrix of the optimized human body structure based on the initial state, constructing an action sequence of the optimized human body structure using the state transition probability matrix, and performing action segmentation on the action sequence to obtain action subunits; The motion capture module is used to identify the motion features of the motion sub-unit, construct the motion label of the motion feature, match the motion label with the preconfigured motion library to obtain the matching action, calculate the probability value of the matching action, determine the target action based on the probability value, and implement motion capture of the animation character to be captured based on the target action.

Images