Animation character motion capture system based on artificial intelligence

Through an animation character motion capture system based on artificial intelligence, multi-dimensional evaluation of action effects and graphic area value encryption is used, the quality and data security issues of animation products in the existing technology are solved, the quality and production efficiency of animation products are improved, and the quality and production efficiency of animation products are reduced. Data attack risk.

CN118587339BActive Publication Date: 2024-11-12ANHUI SHENGZI TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411068020.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-12
Estimated Expiration
2044-08-06

AI Technical Summary

Technical Problem

The existing animation character motion capture system has shortcomings in synthesis effect evaluation and data encryption, which makes it difficult to guarantee the quality and production efficiency of animation products, and has high data security risks.

Method used

Animated character motion capture system based on artificial intelligence is adopted to evaluate the action effects in multiple dimensions through action data acquisition, artificial intelligence processing, action synthesis and additional modules, and protect the data using graphic area values ​​and square models encryption methods.

Benefits of technology

The quality and production efficiency of animation products are improved, and the best neural network model is selected through multi-dimensional indicators to evaluate the action effect, reducing the risk of data being maliciously attacked.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118587339B_ABST
    Figure CN118587339B_ABST
Patent Text Reader

Abstract

The present invention discloses an animation character motion capture system based on artificial intelligence. The present invention evaluates the motion effect by comprehensively evaluating the accuracy index, the fluency index, the fidelity index and the expert valuation multi-dimensional index, so as to obtain the preferred index of the motion effect corresponding to each neural network model. In the calculation process of the preferred index, the worst allowed accuracy index, the worst allowed fluency index, the passing reference fidelity index and the expert's lowest allowed valuation corresponding to each index are added, so as to improve the accuracy of the calculation result. Based on the preferred index of the motion effect corresponding to each neural network model, the quality of the motion synthesis can be comprehensively judged, and the neural network model that meets the animation requirements and the best expectation is selected according to the judgment result, so as to improve the quality and production efficiency of the final animation product.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of animation character capture, and in particular to an animation character motion capture system based on artificial intelligence. Background Art

[0002] As a key means to achieve realistic animation effects, animation character motion capture technology has been widely used in movies, games, virtual reality and other fields.

[0003] However, the animation character motion capture system in the prior art still has the following deficiencies:

[0004] By training the neural network model to learn the mapping relationship between action features and animation character actions, the automatic capture and synthesis of actions can be achieved. Although the above method has completed the automatic capture and synthesis of actions, it does not conduct a comprehensive evaluation of the synthesis effect to ensure the selection of the best neural network model for application, resulting in the quality and production efficiency of animation products cannot be guaranteed;

[0005] Most of the synthetic motion data are encrypted through traditional encryption methods, resulting in a high risk of malicious attacks.

[0006] To this end, an artificial intelligence-based animation character motion capture system was launched. Summary of the invention

[0007] In view of this, the present invention provides an animation character motion capture system based on artificial intelligence to solve the problems raised by the above background technology.

[0008] The purpose of the present invention can be achieved by the following technical solution: An animation character motion capture system based on artificial intelligence, comprising:

[0009] Motion data acquisition module: used to capture the target person's motion data and send it to the artificial intelligence processing module;

[0010] Artificial intelligence processing module: including data preprocessing unit, feature extraction unit and neural network training unit;

[0011] The data preprocessing unit is used to receive the action data of the target person and perform preprocessing;

[0012] The feature extraction unit is used to extract key features from the action data;

[0013] The neural network training unit is used to train each group of neural network models, learn the mapping relationship between action features and animation character actions, and thereby generate action models corresponding to each group of neural network models; each neural network model is represented by a number i, where i=1,2...k, and k is the total number of trained neural network models;

[0014] Action synthesis module: Apply each group of action models generated by the neural network training unit to the animation character to achieve action synthesis, evaluate the effects of each group of synthesized actions, and select the best neural network model for application based on the results of the effect evaluation;

[0015] Additional modules: including data storage management unit and hardware interface unit;

[0016] The data storage management unit is used to classify the stored action data based on the generation time point, create an action data packet corresponding to the time point, compress the action data packet, and then encrypt the compressed action data packet according to a preset encryption rule, and provide data retrieval and management functions;

[0017] The hardware interface unit is used to communicate with various motion capture hardware devices to ensure accurate data transmission;

[0018] User interaction module: provides an operation interface, allowing users to input instructions, adjust parameters, and preview action effects.

[0019] In some embodiments, the effects of the actions composed of each group are evaluated, specifically:

[0020] Determine relevant indicators for evaluating action effectiveness, including accuracy, fluency, realism, and expert evaluation.

[0021] In some embodiments, the effects of the actions composed of each group are evaluated, specifically:

[0022] Extract key points from the original and synthesized motions of the target person;

[0023] For each extracted key point, calculate its angle in three-dimensional space;

[0024] The angle error between the original motion and the synthesized motion of the target person is calculated using the mean square error;

[0025] For each time point, the angle error between the key points of the original action and the synthesized action is calculated; then the angle error between the original action and the synthesized action at each time point is accumulated to obtain the total error of the corresponding key point in the entire action sequence;

[0026] Preset the weight coefficients of different key points, multiply the total errors of different key points by the corresponding weight coefficients, and then sum them up to get the accuracy index Za of the action effect;

[0027] Calculate the velocity curve of each key point, that is, the derivative of the key point position with respect to time, calculate the acceleration curve of the key point, that is, the derivative of the velocity with respect to time, smooth the velocity and acceleration curves, and detect the mutation points in the derivatives of the velocity and acceleration curves; count the number of mutation points, and calculate the density of mutation points in the time series;

[0028] The weight coefficients corresponding to the number of mutation points and the density of mutation points are preset, the number of mutation points and the density of mutation points are multiplied by the corresponding weight coefficients respectively, and then the sum is obtained to obtain the smoothness index Zb of the action effect.

[0029] In some embodiments, evaluating the effects of the actions composed of each group further includes:

[0030] Ensure that the original action data and the synthesized action data of the target person are aligned in time series, and after alignment, construct feature vectors from each key point data; select a similarity measurement method and calculate the similarity of each pair of key point feature vectors of the original action and the synthesized action;

[0031] And based on the preset transformation rules, the key point feature vector similarity of each pair of original actions and synthesized actions is converted into a score; each group of value ranges of the similarity of each key point feature vector is preset, each group of value ranges of the similarity of each key point feature vector is set to correspond to a score, and the key point feature vector similarity of each pair of original actions and synthesized actions is matched with the corresponding group of value ranges, so as to obtain the similarity score of each key point;

[0032] The similarity scores of all key points are averaged to obtain the fidelity index Zc of the action effect.

[0033] In some embodiments, the effects of the actions composed of each group are evaluated, specifically:

[0034] Send the synthetic motion of the target person to a unified platform, and send a review signal to the corresponding account of each motion capture expert on the platform;

[0035] After receiving the review signal, each motion capture expert will evaluate the synthesized motion and score the effect of the synthesized motion;

[0036] After the scoring is completed, the influence coefficient corresponding to each expert's score is obtained based on the authoritative valuation Qw of each expert; then the score of each expert is multiplied by the corresponding influence coefficient, and then added, and the added value is divided by the total number of experts to obtain the expert valuation Zd of the action effect.

[0037] In some embodiments, based on the authoritative valuation Qw of each expert, the influence coefficient corresponding to each expert's score is obtained, specifically:

[0038] Obtain the years of experience of each expert and record it as Na, count the number of times each expert has been evaluated and record it as Nb; preset the weight coefficients of years of experience and number of times evaluated, multiply the years of experience and number of times evaluated of each expert by the corresponding weight coefficient respectively, and then sum them up to obtain the authoritative valuation Qw of each expert;

[0039] Preset the range of each group of valuation values ​​of the authoritative valuation Qw, and set an influence coefficient corresponding to each group of valuation value range;

[0040] The authoritative valuation Qw of each expert is matched with the preset value ranges of each group to obtain the influence coefficient of each expert corresponding to the current time point.

[0041] In some embodiments, the best neural network model is selected based on the results of the effect evaluation, specifically:

[0042] Extract the accuracy index Za, fluency index Zb, fidelity index Zc and expert valuation Zd of the action effect, and number them based on the corresponding applied neural network model to obtain the accuracy index Zai, fluency index Zbi, fidelity index Zci and expert valuation Zdi of the action effect corresponding to each neural network model;

[0043] According to the formula , the accuracy index Zai, fluency index Zbi, fidelity index Zci and expert valuation Zdi of the action effect corresponding to each neural network model are weightedly calculated to obtain the optimal index Yxi of the action effect corresponding to each neural network model; , , as well as They represent the worst allowed accuracy index, the worst allowed fluency index, the passing reference fidelity index, and the expert’s lowest allowed estimate, respectively; , , , They are the influence weight factors of accuracy index Zai, fluency index Zbi, fidelity index Zci and expert valuation Zdi;

[0044] The preferred index Yxi of the action effect corresponding to each neural network model is sorted from large to small, and the neural network model with the largest preferred index Yxi is selected as the best neural network model and applied.

[0045] In some embodiments, the compressed action data packet is encrypted according to a preset encryption rule, specifically:

[0046] Randomly generate a string of codes within a set number range for each group of action data packets to obtain a code sequence for each group of action data packets; the code is composed of numbers, symbols and letters;

[0047] Extract each symbol and letter in each group of action data packet code sequences, and preset each symbol and letter to correspond to a positive integer; convert each symbol and letter in each group of action data packet code sequences into a positive integer according to a preset rule, and add the converted groups of positive integers to obtain the graphic area value of each group of action data packet code sequences;

[0048] The image area value of each group of action data packet code sequence is used as the area value of the square, and a square model corresponding to each group of action data packet code sequence is constructed;

[0049] After the construction is completed, the number of numbers in the code sequence of each group of action data packets after removing symbols and letters is counted as the division ratio, and the corresponding division ratio is performed in the corresponding constructed square model. After the division is completed, each group of numbers is calculated using the preset addition, subtraction, multiplication and division formulas to obtain a series of new numbers corresponding to each group of action data packets, and the new numbers are filled in the square boxes of the divided squares in turn to obtain the encrypted graphics of each group of action data packets.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The present invention evaluates the action effect by comprehensively evaluating the accuracy index, fluency index, fidelity index and expert valuation multi-dimensional indicators, thereby obtaining the preferred index of the action effect corresponding to each neural network model. In the calculation process of the preferred index, the worst allowable accuracy index, the worst allowable fluency index, the passing reference fidelity index and the expert's lowest allowable valuation corresponding to each index are added, thereby improving the accuracy of the calculation result. Based on the preferred index of the action effect corresponding to each neural network model, the quality of the action synthesis can be comprehensively judged, and the neural network model that meets the animation requirements and the best expectations is selected according to the judgment result, thereby improving the quality and production efficiency of the final animation product.

[0052] The present invention reflects the expert's experience and familiarity by combining the expert's score with the impact coefficient corresponding to his authoritative valuation, making the evaluation result more professional and reliable;

[0053] The present invention provides a novel data protection mechanism by adopting an encryption method based on graphic area value and square model, thereby reducing the risk of data being maliciously attacked. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Further details, features and advantages of the present application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0055] Figure 1 It is a principle block diagram of the present invention;

[0056] Figure 2 It is a schematic diagram of the encrypted graph in the present invention. DETAILED DESCRIPTION

[0057] Several embodiments of the present application will be described in more detail below with reference to the accompanying drawings so that those skilled in the art can implement the present application. The present application can be embodied in many different forms and purposes and should not be limited to the embodiments described herein. These embodiments are provided to make the present application comprehensive and complete, and to fully convey the scope of the present application to those skilled in the art. The embodiments do not limit the present application.

[0058] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which this application belongs. It will be further understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the relevant art and / or the context of this specification, and will not be interpreted in an idealized or overly formal sense unless explicitly defined as such herein.

[0059] See also Figure 1-Figure 2 As shown, the animation character motion capture system based on artificial intelligence includes a motion data acquisition module, an artificial intelligence processing module, a motion synthesis module, an additional module and a user interaction module;

[0060] The motion data acquisition module is used to capture the motion data of the target person and send it to the artificial intelligence processing module; including using an optical system, an inertial sensor or a depth camera for capture;

[0061] Obtain the target person's physical movements to provide raw data for subsequent data processing and movement synthesis;

[0062] Optical system: Use markers and high-speed cameras to capture motion, and extract motion data through computer vision technology;

[0063] Inertial sensors: sensors worn on actors capture motion data, including acceleration and rotation information;

[0064] Depth camera: uses depth perception technology to capture movements in three-dimensional space and provide high-precision movement data;

[0065] The artificial intelligence processing module includes a data preprocessing unit, a feature extraction unit, and a neural network training unit;

[0066] The data preprocessing unit is used to receive the target person's motion data and perform preprocessing; preprocessing includes cleaning and standardizing the captured motion data. This includes removing noise, adjusting the data format, etc. to ensure data consistency and availability;

[0067] The feature extraction unit is used to extract key features from the motion data; key features include joint angles, speeds, and accelerations, which are the basis for motion recognition and synthesis;

[0068] The neural network training unit is used to train each group of neural network models, learn the mapping relationship between action features and animation character actions, and thereby generate action models corresponding to each group of neural network models; use deep learning algorithms, such as convolutional neural networks, recurrent neural networks, and long short-term memory networks, to train models to simulate and predict actions; each neural network model is represented by a number i, where i=1,2...k, and k is the total number of trained neural network models;

[0069] The action synthesis module is used to apply each group of action models generated by the neural network training unit to the animation character to realize action synthesis, and evaluate the effects of each group of synthesized actions, and select the best neural network model for application based on the results of the effect evaluation;

[0070] Evaluate the effects of each combined action, specifically:

[0071] First, determine the relevant indicators for evaluating the action effect, including accuracy, fluency, realism and expert evaluation;

[0072] Extract key points from the original and synthesized actions of the target person; key points such as joint positions are set by technicians;

[0073] For each extracted key point, calculate its angle in three-dimensional space; use the rotation representation method of Euler angles;

[0074] Calculate the angle error between the original motion and the synthesized motion of the target person;

[0075] Using common error metrics, such as:

[0076] Mean squared error (MSE): Calculates the average of the squared differences between the original and synthesized joint positions;

[0077] Mean absolute error (MAE): Calculate the average of the absolute values ​​of the differences between the original and synthesized joint positions;

[0078] Root mean square error (RMSE): The square root of the MSE, with the same units as the original data;

[0079] For each time point, calculate the angle error between the original action and the synthesized action at each key point; for each key point, calculate the angle difference between the original action and the synthesized action. This can be achieved by taking the difference of the two angles and normalizing them to the range [-π, π] to evaluate the change of the action over time;

[0080] Then, the angular errors between the original action and the synthesized action at each time point are accumulated to obtain the total error of the corresponding key point in the entire action sequence;

[0081] Preset the weight coefficients of different key points, multiply the total errors of different key points by the corresponding weight coefficients, and then sum them up to get the accuracy index Za of the action effect;

[0082] Calculate the velocity curve of each key point, that is, the derivative of the key point position with respect to time, calculate the acceleration curve of the key point, that is, the derivative of the velocity with respect to time, use a smoothing filter to smooth the velocity and acceleration curves, and then detect the mutation points in the derivatives of the velocity and acceleration curves; the mutation points can be expressed as spikes or discontinuous points in the derivative curve; count the number of mutation points, and calculate the density of mutation points in the time series;

[0083] Preset the weight coefficients corresponding to the number of mutation points and the density of mutation points, multiply the number of mutation points and the density of mutation points by the corresponding weight coefficients respectively, and then sum them up to obtain the smoothness index Zb of the action effect;

[0084] Ensure that the original action data and the synthesized action data of the target person are aligned in time series, and after alignment, construct feature vectors from each key point data; including but not limited to position, speed, acceleration, etc.; select appropriate similarity measurement methods, such as Euclidean distance, Manhattan distance and cosine similarity; calculate the similarity of each pair of key point feature vectors of the original action and the synthesized action;

[0085] And based on the preset transformation rules, the key point feature vector similarity of each pair of original actions and synthesized actions is converted into a score; each group of value ranges of the similarity of each key point feature vector is preset, each group of value ranges of the similarity of each key point feature vector is set to correspond to a score, and the key point feature vector similarity of each pair of original actions and synthesized actions is matched with the corresponding group of value ranges, so as to obtain the similarity score of each key point;

[0086] The scoring range can be set from 1 to 5. The specific conversion rules are preset by the technical staff and can be modified later according to the actual application situation;

[0087] The similarity scores of all key points are averaged to obtain the fidelity index Zc of the action effect;

[0088] Send the target character's synthetic motion to a unified platform, and send review signals to the corresponding accounts of each motion capture expert on the platform; select an evaluation tool that enables experts to watch the animation on a unified platform and record their evaluation;

[0089] After receiving the review signal, each motion capture expert will evaluate the synthesized motion and score the effect of the synthesized motion; the score range is set from 1 to 100, and the higher the score, the better the effect;

[0090] After the scoring is completed, the influence coefficient corresponding to each expert's score is obtained based on the authoritative valuation Qw of each expert; then the score of each expert is multiplied by the corresponding influence coefficient, and then added, and the added value is divided by the total number of experts to obtain the expert valuation Zd of the action effect;

[0091] Extract the accuracy index Za, fluency index Zb, fidelity index Zc and expert valuation Zd of the action effect, and number them based on the corresponding applied neural network model to obtain the accuracy index Zai, fluency index Zbi, fidelity index Zci and expert valuation Zdi of the action effect corresponding to each neural network model;

[0092] According to the formula , the accuracy index Zai, fluency index Zbi, fidelity index Zci and expert valuation Zdi of the action effect corresponding to each neural network model are weightedly calculated to obtain the optimal index Yxi of the action effect corresponding to each neural network model; , , as well as They represent the worst allowed accuracy index, the worst allowed fluency index, the passing reference fidelity index, and the expert’s lowest allowed estimate, respectively; , , , They are the influence weight factors of accuracy index Zai, fluency index Zbi, fidelity index Zci and expert valuation Zdi;

[0093] The preferred index Yxi of the action effect corresponding to each neural network model is sorted from large to small, and the neural network model with the largest preferred index Yxi is used as the best neural network model;

[0094] By comprehensively evaluating the above parameters, we can make a comprehensive judgment on the quality of motion synthesis, and select the neural network model that best meets animation requirements and expectations based on the judgment results, thereby improving the quality and production efficiency of the final animation product.

[0095] Get the authoritative valuation Qw of each expert, specifically:

[0096] Obtain the years of experience of each expert and record it as Na, and count the number of evaluations of each expert and record it as Nb;

[0097] Reflect the expert's experience and familiarity based on their years of practice and number of assessments;

[0098] Preset the weight coefficients of years of experience and times of evaluation, multiply the years of experience and times of evaluation of each expert by the corresponding weight coefficient, and then sum them up to get the authoritative valuation Qw of each expert;

[0099] Preset the range of each group of valuation values ​​of the authoritative valuation Qw, and set each group of valuation value range to correspond to an influence coefficient; the larger the authoritative valuation Qw, the larger the corresponding matching influence coefficient, and the influence coefficient value range is set to 1.35-1.72;

[0100] Match the authoritative valuation Qw of each expert with the preset value ranges of each group to obtain the influence coefficient of each expert at the current time point;

[0101] The additional module includes a data storage management unit and a hardware interface unit;

[0102] The data storage management unit is used to classify the stored action data based on the generation time point, create an action data packet corresponding to the time point, compress the action data packet, and then encrypt the compressed action data packet according to the preset encryption rules, and provide data retrieval and management functions; the preset encryption rules include;

[0103] Use lossless compression algorithms, such as DEFLATE, LZMA, or Brotli, to ensure that the compressed data can be fully recovered;

[0104] The compressed action data packet is encrypted according to the preset encryption rules, specifically:

[0105] Randomly generate a string of codes within a set number range for each group of action data packets to obtain a code sequence for each group of action data packets; the code is composed of numbers, symbols and letters;

[0106] Extract each symbol and letter from each group of action data packet code sequences, and preset each symbol and letter to correspond to a positive integer; for example, p=1, ~=3; convert each symbol and letter in each group of action data packet code sequences into a positive integer according to a preset rule, and add the converted groups of positive integers to obtain the graphic area value of each group of action data packet code sequences;

[0107] The image area value of each group of action data packet code sequence is used as the area value of the square, and a square model corresponding to each group of action data packet code sequence is constructed;

[0108] After the construction is completed, the number of numbers in the code sequence of each group of action data packets after removing symbols and letters is counted as the division ratio, and the corresponding division ratio is divided in the corresponding square model. After the division is completed, each group of numbers is calculated using the preset addition, subtraction, multiplication and division formulas to obtain a series of new numbers corresponding to each group of action data packets, and the new numbers are filled into the boxes of the divided squares in sequence to obtain the encrypted graphics of each group of action data packets;

[0109] Preset addition, subtraction, multiplication and division formulas such as [gf+3)×gr-1] / 2, where gf represents the number to be substituted into the calculation and gr is the preset additional value, with a range of 1.5-1.9;

[0110] The method using the area value of the graph and the square model provides an unconventional encryption method, which may reduce the risk of malicious attacks.

[0111] It should be noted that, for example, if a series of new numbers are obtained, such as 3, 6, 11, 15, 9, 7, 2, 10, and 1, the constructed square model is divided by dotted lines according to the division ratio, and the series of numbers obtained are filled in sequentially;

[0112] Mapping the data packet to be encrypted into the matrix of the encryption graph is achieved by converting the data packet into a digital sequence and performing corresponding transformations based on the values ​​in the encryption graph.

[0113] Fill the transformed data points into the corresponding positions of the encrypted graph. For example, if the data point is 3, then fill it into the corresponding number 3 of the encrypted graph. If the data point is 6, then fill it into the corresponding number 6 of the encrypted graph.

[0114] To decrypt the data, the receiver needs to have the same encryption graph and transformation rules and perform the operations in reverse order to convert the encrypted data back to the original data.

[0115] The hardware interface unit is used to communicate with various motion capture hardware devices to ensure accurate data transmission;

[0116] The user interaction module is used to provide an operation interface, allowing users to input instructions, adjust parameters, and preview action effects;

[0117] The above formulas are obtained by collecting a large amount of data and performing software simulation, and a formula close to the actual value is selected. The influencing weight factor and specific coefficient value in the formula are set by technical personnel in this field according to actual conditions, and can be adjusted and modified later.

[0118] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. Animation character motion capture system based on artificial intelligence, characterized by: include: Motion data acquisition module: used to capture the target person's motion data and send it to the artificial intelligence processing module; Artificial intelligence processing module: including data preprocessing unit, feature extraction unit and neural network training unit; The data preprocessing unit is used to receive the action data of the target person and perform preprocessing; The feature extraction unit is used to extract key features from the action data; The neural network training unit is used to train each group of neural network models, learn the mapping relationship between action features and animation character actions, and thereby generate action models corresponding to each group of neural network models; each neural network model is represented by a number i, where i=1,2...k, and k is the total number of trained neural network models; Action synthesis module: Apply each group of action models generated by the neural network training unit to the animation character to achieve action synthesis, evaluate the effects of each group of synthesized actions, and select the best neural network model for application based on the results of the effect evaluation; Additional modules: including data storage management unit and hardware interface unit; The data storage management unit is used to classify the stored action data based on the generation time point, create an action data packet corresponding to the time point, compress the action data packet, and then encrypt the compressed action data packet according to a preset encryption rule, and provide data retrieval and management functions; The compressed action data packet is encrypted according to the preset encryption rules, specifically: Randomly generate a string of codes within a set number range for each group of action data packets to obtain a code sequence for each group of action data packets; the code is composed of numbers, symbols and letters; Extract each symbol and letter in each group of action data packet code sequences, and preset each symbol and letter to correspond to a positive integer; convert each symbol and letter in each group of action data packet code sequences into a positive integer according to a preset rule, and add the converted groups of positive integers to obtain the graphic area value of each group of action data packet code sequences; The image area value of each group of action data packet code sequence is used as the area value of the square, and a square model corresponding to each group of action data packet code sequence is constructed; After the construction is completed, the number of numbers in the code sequence of each group of action data packets after removing symbols and letters is counted as the division ratio, and the corresponding division ratio is divided in the corresponding square model. After the division is completed, each group of numbers is calculated using the preset addition, subtraction, multiplication and division formulas to obtain a series of new numbers corresponding to each group of action data packets, and the new numbers are filled into the boxes of the divided squares in sequence to obtain the encrypted graphics of each group of action data packets; The hardware interface unit is used to communicate with various motion capture hardware devices to ensure accurate data transmission; User interaction module: provides an operation interface, allowing users to input instructions, adjust parameters, and preview action effects.

2. The animation character motion capture system based on artificial intelligence according to claim 1, characterized in that: Evaluate the effects of each combined action, specifically: Determine relevant indicators for evaluating action effectiveness, including accuracy, fluency, realism, and expert evaluation.

3. The animation character motion capture system based on artificial intelligence according to claim 2, characterized in that: Evaluate the effects of each combined action, specifically: Extract key points from the original and synthesized motions of the target person; For each extracted key point, calculate its angle in three-dimensional space; The angle error between the original motion and the synthesized motion of the target person is calculated using the mean square error; For each time point, calculate the angle error between the key points of the original action and the synthetic action; Then, the angular errors between the original action and the synthesized action at each time point are accumulated to obtain the total error of the corresponding key point in the entire action sequence; Preset the weight coefficients of different key points, multiply the total errors of different key points by the corresponding weight coefficients, and then sum them up to get the accuracy index Za of the action effect; Calculate the velocity curve of each key point, that is, the derivative of the key point position with respect to time, calculate the acceleration curve of the key point, that is, the derivative of the velocity with respect to time, smooth the velocity and acceleration curves, and detect the mutation points in the derivatives of the velocity and acceleration curves; count the number of mutation points, and calculate the density of mutation points in the time series; The weight coefficients corresponding to the number of mutation points and the density of mutation points are preset, the number of mutation points and the density of mutation points are multiplied by the corresponding weight coefficients respectively, and then the sum is obtained to obtain the smoothness index Zb of the action effect.

4. The artificial intelligence-based animation character motion capture system according to claim 3, characterized in that: The effectiveness of each combined action is evaluated, including: Ensure that the original action data and the synthesized action data of the target person are aligned in time series, and after alignment, construct feature vectors from each key point data; select a similarity measurement method and calculate the similarity of each pair of key point feature vectors of the original action and the synthesized action; And based on the preset transformation rules, the key point feature vector similarity of each pair of original actions and synthesized actions is converted into a score; each group of value ranges of the similarity of each key point feature vector is preset, each group of value ranges of the similarity of each key point feature vector is set to correspond to a score, and the key point feature vector similarity of each pair of original actions and synthesized actions is matched with the corresponding group of value ranges, so as to obtain the similarity score of each key point; The similarity scores of all key points are averaged to obtain the fidelity index Zc of the action effect.

5. The artificial intelligence-based animation character motion capture system according to claim 4, characterized in that: Evaluate the effects of each combined action, specifically: Send the synthetic motion of the target person to a unified platform, and send a review signal to the corresponding account of each motion capture expert on the platform; After receiving the review signal, each motion capture expert will evaluate the synthesized motion and score the effect of the synthesized motion; After the scoring is completed, the influence coefficient corresponding to each expert's score is obtained based on the authoritative valuation Qw of each expert; then the score of each expert is multiplied by the corresponding influence coefficient, and then added, and the added value is divided by the total number of experts to obtain the expert valuation Zd of the action effect.

6. The artificial intelligence-based animation character motion capture system according to claim 5, characterized in that: Based on the authoritative valuation Qw of each expert, the influence coefficient corresponding to each expert's score is obtained, specifically: Obtain the years of experience of each expert and record it as Na, count the number of times each expert has been evaluated and record it as Nb; preset the weight coefficients of years of experience and number of times evaluated, multiply the years of experience and number of times evaluated of each expert by the corresponding weight coefficient respectively, and then sum them up to obtain the authoritative valuation Qw of each expert; Preset the range of each group of valuation values ​​of the authoritative valuation Qw, and set an influence coefficient corresponding to each group of valuation value range; The authoritative valuation Qw of each expert is matched with the preset value ranges of each group to obtain the influence coefficient of each expert corresponding to the current time point.

7. The artificial intelligence-based animation character motion capture system according to claim 6, characterized in that: The best neural network model is selected based on the results of effect evaluation, specifically: Extract the accuracy index Za, fluency index Zb, fidelity index Zc and expert valuation Zd of the action effect, and number them based on the corresponding applied neural network model to obtain the accuracy index Zai, fluency index Zbi, fidelity index Zci and expert valuation Zdi of the action effect corresponding to each neural network model; According to the formula The accuracy index Zai, fluency index Zbi, fidelity index Zci and expert valuation Zdi of the action effect corresponding to each neural network model are weighted to obtain the optimal index Yxi of the action effect corresponding to each neural network model; 参考 、Zbi 参考 、Zci 参考 and Zdi 参考 They represent the worst allowed accuracy index, the worst allowed fluency index, the passing reference fidelity index, and the expert's lowest allowed valuation respectively; α, β, χ, and η are the influencing weight factors of the accuracy index Zai, the fluency index Zbi, the fidelity index Zci, and the expert valuation Zdi respectively; The preferred index Yxi of the action effect corresponding to each neural network model is sorted from large to small, and the neural network model with the largest preferred index Yxi is selected as the best neural network model and applied.