Virtual character behavior action simulation system and method based on artificial intelligence

Abstract

The invention discloses a virtual character behavior action simulation system and method based on artificial intelligence, and relates to the technical field of artificial intelligence. The system comprises an action data acquisition and analysis module, a behavior instruction analysis and motion sequence generation module, a complementary frame animation generation module, a complementary frame animation quality evaluation and problem identification module and a complementary frame animation optimization and adjustment module. The action data acquisition and analysis module acquires and analyzes action data of a virtual character; the behavior instruction analysis and motion sequence generation module generates a motion sequence of the virtual character according to the behavior instruction; the frame compensation animation generation module generates transition frames between adjacent actions and gathers the transition frames to obtain frame compensation animations; a complementary frame animation quality evaluation and problem identification module performs quality evaluation on the generated complementary frame animation and identifies problem frames; and the frame compensation animation optimization and adjustment module adjusts and optimizes the problem frame according to the evaluation result, and ensures that the final animation quality reaches the standard through secondary quality evaluation.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a virtual character behavior and action simulation system and method based on artificial intelligence. Background Art

[0002] With the rapid development of computer graphics and artificial intelligence technology, virtual character behavior animation has been widely used in interactive entertainment and simulation systems such as movies, games, virtual reality (VR), and augmented reality (AR). In order to ensure the naturalness and smoothness of these virtual character movements, interpolation animation has become an important means to optimize the movement performance of virtual characters. Interpolation animation refers to inserting additional animation frames between two main actions to achieve smooth transitions and more delicate performances, and enhance the realism and coherence of the animation.

[0003] Although some automatic interpolation generation methods and algorithms have been proposed, solutions based on interpolation technology are more common, and these methods can reduce manual intervention to a certain extent when dealing with simple and regular animation tasks. However, animations generated using interpolation technology often have some problems, such as unnatural transitions, inconsistent movement speeds, and distorted movement forms. Moreover, for the task of generating interpolation animations using interpolation technology, existing technologies are usually unable to automatically identify defects in these interpolation animations, let alone make corresponding modification suggestions for the defects in the interpolation animations, which further increases the difficulty and cost of production. Summary of the invention

[0004] The purpose of the present invention is to provide a virtual character behavior and action simulation system and method based on artificial intelligence to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: A method for simulating virtual character behavior based on artificial intelligence, comprising the following steps: Step S100. Acquire action data of the virtual character from the database, analyze the acquired action data of the virtual character, thereby obtaining key frame data of each action of the virtual character, and establish a key frame data set of the virtual character based on the key frame data; Step S200. Obtaining the behavior instruction of the virtual character, searching in the key frame data set based on the behavior instruction of the virtual character, thereby obtaining the motion sequence of the virtual character corresponding to the behavior instruction; generating a transition frame between two adjacent actions in the motion sequence according to the motion sequence of the virtual character, thereby obtaining a frame-filling animation between two actions of the virtual character; Step S300. Obtain motion data corresponding to the interlaced frame animation between two actions of the virtual character, analyze the motion data corresponding to the interlaced frame animation, and thereby identify the focus frame in the interlaced frame animation; perform quality assessment on the focus frame in combination with the motion data; and identify the problem frame in the interlaced frame animation based on the quality assessment result; Step S400. Analyze the problem frames in the interframe animation, generate corresponding adjustment suggestions, and output the problem frames and the corresponding adjustment suggestions to relevant personnel, who then adjust and optimize the interframe animation based on the adjustment suggestions; perform a secondary quality assessment on the adjusted and optimized interframe animation, and perform corresponding processing based on the secondary quality assessment results.

[0006] Furthermore, step S100 includes: S101. Obtaining the action data of the virtual character from the database, and the action data is based on motion capture technology or manually designed animation data sets, each data includes a time series of action data, wherein each time point corresponds to the posture information of each joint or bone of the character (such as the angle and position of the joint, etc.). The action data is composed of a number of frames, each frame represents the posture of the virtual character at a certain moment, and each action data is represented as an action sequence P, and P={p1,p2,...,pn}, wherein p1 represents the posture of the virtual character in the first frame, p2 represents the posture of the virtual character in the second frame, and so on, pn represents the posture of the virtual character in the nth frame, and n represents the number of frames in the action sequence P; S102. Analyze the acquired action data of the virtual character, and identify the key frames of each action of the virtual character, wherein the key frames refer to the action start frame and the action end frame of the virtual character; the key frame identification process is as follows: According to each action sequence of the virtual character, the difference Δpi between the virtual character postures corresponding to frame pi and frame pi+1 is calculated, and Δpi=||(pi+1)-pi||, where ||(pi+1)-pi|| represents the Euclidean distance between frame pi and frame pi+1, which is used to quantify the degree of change of the action of the virtual character between the two frames. If the action has changed significantly, the difference Δpi will be large, otherwise it will be small; summarize the differences Δpi, and draw the corresponding difference change curve Q1, and the horizontal axis of the difference change curve Q1 is the sequential number of the differences Δpi, and the vertical axis of the difference change curve Q1 is the difference Δpi; if a certain section of the difference change curve Q1 is the first If a certain segment of the difference change curve Q1 shows an increasing trend, the size relationship between the vertical axis Δpi of the corresponding difference change curve segment and the threshold ΔTstart is compared. If Δpi is greater than the threshold ΔTstart, the frame pi+1 is marked as the action start frame p_start of the virtual character; if a segment of the difference change curve Q1 shows a decreasing trend for the first time, the size relationship between the vertical axis Δpi of the corresponding difference change curve segment and the threshold ΔTend is compared. If Δpi is less than the threshold ΔTend, the frame pi+1 is marked as the action end frame p_end of the virtual character; for each action sequence of the virtual character, the corresponding action start frame and action end frame are summarized to identify the key frame of the corresponding action sequence; S103. According to the key frame of each action of the identified virtual character, the action data of the virtual character is divided to obtain the key frame data corresponding to the action data of the virtual character, and the key frame data corresponding to each action data of the virtual character are summarized to establish a key frame data set G of the virtual character, and G={g1,g2,...,gm}, wherein g1 represents the key frame data of the first action of the virtual character, g2 represents the key frame data of the second action of the virtual character, and so on, gm represents the key frame data of the mth action of the virtual character, and m represents the action number of the virtual character.

[0007] By analyzing the action data of the virtual character and identifying the key frames, the entire action sequence can be divided into a more streamlined key frame data set, thereby reducing the amount of calculations that need to be processed. By calculating the differences between action frames and using this as a basis for identifying key frames, redundant frames that have little impact on the action performance can be effectively eliminated, improving the performance quality of the virtual character's actions; in the rendering of animation effects, reducing unnecessary intermediate frames helps to reduce the unnatural and unsmooth transition effects. Key frame recognition provides an important reference for subsequent action generation and editing, especially in animation production and game development, where the animation of virtual characters usually relies on key frames; by accurately identifying the start and end frames of the action, it can help animators and developers more accurately control the smoothness and expressiveness of the character's actions.

[0008] Further, step S200 includes: S201. Obtaining a behavior instruction of a virtual character, wherein the behavior instruction of the virtual character is composed of an action number of the corresponding virtual character, and the behavior instruction of the virtual character is represented by B, and B={b1,b2,...,bk}, wherein b1 represents the first action number of the virtual character, b2 represents the second action number of the virtual character, and so on, bk represents the second action number of the virtual character; k represents the total number of action numbers corresponding to the behavior instruction of the virtual character, and k≤m; searching in the key frame data set G according to the behavior instruction of the virtual character, so as to obtain the key frame data sequence corresponding to the behavior instruction of the virtual character. Column, obtain corresponding action data in the database according to the key frame data sequence, mark the key frame position in the action data, filter the action data based on the key frame mark, retain the data from the action start frame to the action end frame for each action data, sort the filtered action data according to the behavior instructions of the virtual character, so as to obtain the motion sequence D of the virtual character corresponding to the behavior instructions, and D={d1,d2,...,dk}, and d1 represents the motion data corresponding to the behavior instruction b1, d2 represents the motion data corresponding to the behavior instruction b2, and so on, dk represents the motion data corresponding to the behavior instruction bk; S202. Perform physical simulation on the motion sequence D of the virtual character to obtain the physical properties of the virtual character at each key frame moment; for two adjacent actions bj and bj+1 in the motion sequence of the virtual character, obtain the action end frame pj_end of bj and the action start frame pj+1_start of bj+1 according to the corresponding key frame data, generate the transition frame pt between the adjacent actions bj and bj+1 by interpolation method, and the corresponding calculation formula is: p=(1-α)·pj_end+α·pj+1_start; wherein α represents the interpolation coefficient, and the value is [0,1]; according to the physical properties of the virtual character at each key frame moment, mark the motion data corresponding to the two adjacent actions bj and bj+1 in the motion sequence of the virtual character, mark the key frames and corresponding physical states of the adjacent actions bj and bj+1 and the generated transition frames, and use these data as training data; through the trained model, input the end frame and start frame of the two adjacent actions, the model will generate the corresponding interpolation frame sequence, combine the interpolation frame sequence with the motion sequence D, and obtain the interpolation frame animation between the two actions of the virtual character.

[0009] Furthermore, step S300 includes: S301. Obtain the interpolation frame animation between two actions of the virtual character, obtain the corresponding motion data of the virtual character for each transition frame p corresponding to the interpolation frame animation, and calculate the motion change ΔMp(t) between two adjacent transition frames, where the specific formula is: ΔMp(t)={Σe∈[1,E],[Mep(t+1)-Mep(t)] 2}^(1 / 2); Where t represents the time point, Mep(t+1) represents the e-th kinematic feature of the transition frame corresponding to the time point t+1, and Mep(t) represents the e-th kinematic feature of the transition frame corresponding to the time point t; E represents the number of kinematic features, and the kinematic features include the joint or bone positions of the virtual character, the changes in joint angles, the rotation center of the skeleton, etc.; the motion changes ΔMp(t) between all adjacent transition frames are summarized, and the motion change curve Q2 is drawn with the time point as the horizontal axis and the motion change ΔMp(t) as the vertical axis, and the tangent slope R corresponding to each data point on the motion change curve Q2 is calculated, and the tangent slope R corresponding to each data point is compared with the set threshold Rth, and the transition frame of the data point whose tangent slope R is greater than the set threshold Rth is marked as the focus frame; S302. Summarize all the focus frames in the interpolation animation between two actions of the virtual character, and for each focus frame, calculate the smoothness index S of the focus frame in combination with the motion change ΔMp(t) corresponding to each transition frame p in the interpolation animation. The corresponding calculation formula is: S(pt)=1 / {1+Σt∈[1,T],ΔMp(t)·β}; Among them, T represents the total number of time points corresponding to the interframe animation between two actions of the virtual character, which is equal to the number of transition frames corresponding to the interframe animation between two actions of the virtual character; β represents the smoothness weight, and it ranges from 0 to 1; the physical consistency score SW corresponding to each focus frame is calculated, and the specific calculation formula is: SW(pt)=1-|θp(t+1)-θp(t)| / (θ_max-θ_min); Among them, θp(t+1) represents the joint angle value of the virtual character in the transition frame corresponding to the time point t+1, θp(t) represents the joint angle value of the virtual character in the transition frame corresponding to the time point t; θ_max-θ_min represents the maximum angle variation range allowed for the joints of the virtual character; the action consistency index C corresponding to the focus frame in the interpolation frame animation between two actions of the virtual character is calculated, and the corresponding calculation formula is: C(pt)={(1 / T)Σt∈[1,T],Σe∈[1,E],||Me(pt)-M(pt)||} / T; Where M(pt) represents the expected motion characteristics of the transition frame corresponding to time point t; S303. According to the smoothness index S, physical consistency score SW and motion consistency index C of the focus frame, calculate the corresponding comprehensive quality score SQ, and SQ(pt)=w1·S(pt)+w2·SW(pt)+w3·C(pt); wherein w1, w2 and w3 are the weight coefficients of the smoothness index S, the physical consistency score SW and the motion consistency index C respectively, and the sum of the weights is 1; compare the comprehensive quality score SQ of the focus frame with the threshold SQ0, and mark the focus frame whose comprehensive quality score SQ is less than the threshold SQ0 as a problem frame.

[0010] Furthermore, step S400 includes: S401. Obtain the problem frame of the interpolation animation between two actions of the virtual character, extract the smoothness index S, physical consistency score SW and action consistency index C corresponding to the problem frame, and compare the smoothness index S, physical consistency score SW and action consistency index C with the corresponding smoothness index threshold S0, physical consistency score threshold SW0 and action consistency index threshold C0 respectively; if the smoothness index S is less than the smoothness index threshold S0, it means that the defect corresponding to the problem frame is that the transition speed is too fast or too slow, and the improvement suggestion is "optimize the interpolation algorithm or increase or decrease the number of interpolation frames"; if the physical consistency score SW is less than the physical consistency score threshold SW0, it means that the defect corresponding to the problem frame is a violation of the physical law, and the improvement suggestion is "adjust the physical properties of the interpolation animation"; if the action consistency index C is greater than the action consistency index threshold C0, it means that the defect corresponding to the problem frame is inconsistent behavior of the virtual character, and the improvement suggestion is "adjust the transition logic between actions, and optimize the selection and sorting of action sequences"; S402. Output the problem frames and the corresponding adjustment suggestions to relevant personnel, who adjust and optimize the interpolation frame animation based on the adjustment suggestions; and perform a secondary quality assessment on the adjusted interpolation frame animation, obtain the comprehensive quality score SQ1 corresponding to the secondary quality assessment, and compare the comprehensive quality score SQ1 with the threshold value SQ0; if the comprehensive quality score SQ1 is less than the threshold value SQ0, repeat the content of S401 until the corresponding comprehensive quality score is greater than or equal to the threshold value SQ0.

[0011] A virtual character behavior and motion simulation system based on artificial intelligence, the system includes: an action data acquisition and analysis module, a behavior instruction parsing and motion sequence generation module, a frame-filling animation generation module, a frame-filling animation quality assessment and problem identification module, and a frame-filling animation optimization and adjustment module; The action data acquisition and analysis module acquires the action data of the virtual character from the database, analyzes the acquired action data of the virtual character, thereby obtaining key frame data of each action of the virtual character, and establishes a key frame data set of the virtual character based on the key frame data; The behavior instruction parsing and motion sequence generation module obtains the behavior instruction of the virtual character, searches in the key frame data set based on the behavior instruction of the virtual character, and thus obtains the motion sequence of the virtual character corresponding to the behavior instruction; The interpolation frame animation generation module generates a transition frame between two adjacent actions in the motion sequence according to the motion sequence of the virtual character, thereby obtaining an interpolation frame animation between two actions of the virtual character; The interframe animation quality assessment and problem identification module obtains motion data corresponding to the interframe animation between two actions of the virtual character, analyzes the motion data corresponding to the interframe animation, and thus identifies the focus frames in the interframe animation; performs quality assessment on the focus frames in combination with the motion data; and identifies the problem frames in the interframe animation based on the quality assessment results; The frame-interpolation animation optimization and adjustment module analyzes the problem frames in the frame-interpolation animation, generates corresponding adjustment suggestions, and outputs the problem frames and corresponding adjustment suggestions to relevant personnel, who then adjust and optimize the frame-interpolation animation based on the adjustment suggestions; conducts a secondary quality assessment on the adjusted and optimized frame-interpolation animation, and performs corresponding processing based on the secondary quality assessment results.

[0012] Further, the motion data acquisition and analysis module includes a motion data acquisition unit, a motion data analysis unit, and a key frame data set construction unit; The motion data acquisition unit acquires the motion data of the virtual character from the database; the motion data analysis unit analyzes the acquired motion data, calculates the difference between the motion frames, draws the difference change curve, and identifies the key frame of each action according to the threshold; the key frame data set construction unit establishes the key frame data set of the virtual character based on the key frame data.

[0013] Further, the behavior instruction parsing and motion sequence generation module includes a behavior instruction acquisition unit and a motion sequence generation unit; The behavior instruction acquisition unit receives the behavior instruction of the virtual character; the motion sequence generation unit searches and obtains the corresponding key frame data sequence in the key frame data set according to the behavior instruction, and generates the motion sequence of the virtual character through screening and sorting.

[0014] The interpolation frame animation generation module includes a physical simulation unit, a transition frame generation unit and an interpolation frame animation generation unit; The physical simulation unit simulates the physical properties of the virtual character at each key frame moment, thereby obtaining the physical properties of the virtual character at each key frame moment; the transition frame generation unit generates a transition frame between two actions based on the motion sequence and physical simulation of the virtual character; the infill animation generation unit summarizes the transition frames between two adjacent actions of the virtual character, thereby generating an infill animation.

[0015] Furthermore, the frame-filling animation quality assessment and problem identification module includes a motion variation calculation unit, a focus frame identification unit, and a problem frame annotation unit; The motion change calculation unit calculates the motion change between each transition frame in the frame-complementing animation; the focus frame identification unit identifies the focus frame according to the motion change; the problem frame labeling unit calculates the comprehensive quality score of the focus frame, identifies and labels the problem frame. The focus frame identification unit identifies the focus frame according to the motion change; the problem frame labeling unit calculates the comprehensive quality score of the focus frame, identifies the problem frame according to the comprehensive quality score, and labels the problem frame.

[0016] Furthermore, the frame-filling animation optimization and adjustment module includes a problem frame analysis unit, an optimization suggestion generation unit, and a secondary quality assessment unit; The problem frame analysis unit analyzes the motion data of the problem frame and identifies the corresponding defects; the optimization suggestion generation unit generates corresponding adjustment suggestions to relevant personnel based on the analysis results of the problem frame analysis unit, and the relevant personnel make corresponding optimization adjustments; the secondary quality assessment unit performs a secondary quality assessment on the adjusted frame-filling animation and performs corresponding processing based on the secondary quality assessment results.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention automatically identifies defects in the interpolation animation through an artificial intelligence-based method, and generates specific adjustment suggestions based on the defects, which greatly reduces the need for manual intervention; the existing interpolation method can usually only generate interpolation animation, and cannot effectively evaluate and optimize the generated animation, while the present invention can evaluate the quality in real time during the interpolation process, ensuring that the generated animation is more natural, smooth, and in line with physical laws. The present invention calculates the motion change between adjacent transition frames and identifies the focus frame through tangent slope analysis, and further conducts a comprehensive quality evaluation of the interpolation animation through smoothness index, physical consistency score and action consistency index; this method can accurately detect unnatural and abrupt changes in transition animation, locate problem frames in time, and provide a clear direction for further optimizing the interpolation animation. After identifying the problem frame, the present invention gives specific adjustment suggestions based on the smoothness, physical consistency and action consistency of the problem frame. For example, if the smoothness of some problem frames is not good, the interpolation algorithm can be optimized or the number of interpolation frames can be adjusted; if there is a violation of physical laws, the physical properties need to be adjusted; if the character behavior is inconsistent, the transition logic between actions needs to be optimized. Through this intelligent adjustment mechanism, the dependence on manual adjustment in the production process is reduced, and the production cost and time are reduced. The present invention also includes a step of performing a secondary quality evaluation on the adjusted and optimized interpolated frame animation to ensure that the adjusted animation reaches the ideal quality standard. If the secondary evaluation result fails to meet the preset quality threshold, the present invention will perform optimization in a loop until the generated interpolated frame animation meets the requirements. This closed-loop optimization process can ensure that the final animation quality is optimal. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 It is a schematic diagram of a module of a virtual character behavior and action simulation system based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] See also Figure 1 , the present invention provides a technical solution: A virtual character behavior and motion simulation system based on artificial intelligence, the system includes: an action data acquisition and analysis module, a behavior instruction parsing and motion sequence generation module, a frame-filling animation generation module, a frame-filling animation quality assessment and problem identification module, and a frame-filling animation optimization and adjustment module; The action data acquisition and analysis module acquires the action data of the virtual character from the database, analyzes the acquired action data of the virtual character, thereby obtaining key frame data of each action of the virtual character, and establishes a key frame data set of the virtual character based on the key frame data; The behavior instruction parsing and motion sequence generation module obtains the behavior instruction of the virtual character, searches in the key frame data set based on the behavior instruction of the virtual character, and thus obtains the motion sequence of the virtual character corresponding to the behavior instruction; The interpolation frame animation generation module generates a transition frame between two adjacent actions in the motion sequence according to the motion sequence of the virtual character, thereby obtaining an interpolation frame animation between two actions of the virtual character; The interframe animation quality assessment and problem identification module obtains motion data corresponding to the interframe animation between two actions of the virtual character, analyzes the motion data corresponding to the interframe animation, and thus identifies the focus frames in the interframe animation; performs quality assessment on the focus frames in combination with the motion data; and identifies the problem frames in the interframe animation based on the quality assessment results; The frame-interpolation animation optimization and adjustment module analyzes the problem frames in the frame-interpolation animation, generates corresponding adjustment suggestions, and outputs the problem frames and corresponding adjustment suggestions to relevant personnel, who then adjust and optimize the frame-interpolation animation based on the adjustment suggestions; conducts a secondary quality assessment on the adjusted and optimized frame-interpolation animation, and performs corresponding processing based on the secondary quality assessment results.

[0021] The motion data acquisition and analysis module includes a motion data acquisition unit, a motion data analysis unit and a key frame data set construction unit; The motion data acquisition unit acquires the motion data of the virtual character from the database; the motion data analysis unit analyzes the acquired motion data, calculates the difference between the motion frames, draws the difference change curve, and identifies the key frame of each action according to the threshold; the key frame data set construction unit establishes the key frame data set of the virtual character based on the key frame data.

[0022] The behavior instruction parsing and motion sequence generation module includes a behavior instruction acquisition unit and a motion sequence generation unit; The behavior instruction acquisition unit receives the behavior instruction of the virtual character; the motion sequence generation unit searches and obtains the corresponding key frame data sequence in the key frame data set according to the behavior instruction, and generates the motion sequence of the virtual character through screening and sorting.

[0023] The interpolation frame animation generation module includes a physical simulation unit, a transition frame generation unit and an interpolation frame animation generation unit; The physical simulation unit simulates the physical properties of the virtual character at each key frame moment, thereby obtaining the physical properties of the virtual character at each key frame moment; the transition frame generation unit generates a transition frame between two actions based on the motion sequence and physical simulation of the virtual character; the infill animation generation unit summarizes the transition frames between two adjacent actions of the virtual character, thereby generating an infill animation.

[0024] The frame-filling animation quality assessment and problem identification module includes a motion variation calculation unit, a focus frame identification unit, and a problem frame annotation unit; The motion change calculation unit calculates the motion change between each transition frame in the frame-complementing animation; the focus frame identification unit identifies the focus frame according to the motion change; the problem frame labeling unit calculates the comprehensive quality score of the focus frame, identifies and labels the problem frame. The focus frame identification unit identifies the focus frame according to the motion change; the problem frame labeling unit calculates the comprehensive quality score of the focus frame, identifies the problem frame according to the comprehensive quality score, and labels the problem frame.

[0025] The frame-filling animation optimization and adjustment module includes a problem frame analysis unit, an optimization suggestion generation unit, and a secondary quality assessment unit; The problem frame analysis unit analyzes the motion data of the problem frame and identifies the corresponding defects; the optimization suggestion generation unit generates corresponding adjustment suggestions to relevant personnel based on the analysis results of the problem frame analysis unit, and the relevant personnel make corresponding optimization adjustments; the secondary quality assessment unit performs a secondary quality assessment on the adjusted frame-filling animation and performs corresponding processing based on the secondary quality assessment results.

[0026] A method for simulating virtual character behavior based on artificial intelligence, comprising the following steps: Step S100. Acquire action data of the virtual character from the database, analyze the acquired action data of the virtual character, thereby obtaining key frame data of each action of the virtual character, and establish a key frame data set of the virtual character based on the key frame data; Step S200. Obtaining the behavior instruction of the virtual character, searching in the key frame data set based on the behavior instruction of the virtual character, thereby obtaining the motion sequence of the virtual character corresponding to the behavior instruction; generating a transition frame between two adjacent actions in the motion sequence according to the motion sequence of the virtual character, thereby obtaining a frame-filling animation between two actions of the virtual character; Step S300. Obtain motion data corresponding to the interlaced frame animation between two actions of the virtual character, analyze the motion data corresponding to the interlaced frame animation, and thereby identify the focus frame in the interlaced frame animation; perform quality assessment on the focus frame in combination with the motion data; and identify the problem frame in the interlaced frame animation based on the quality assessment result; Step S400. Analyze the problem frames in the interframe animation, generate corresponding adjustment suggestions, and output the problem frames and the corresponding adjustment suggestions to relevant personnel, who then adjust and optimize the interframe animation based on the adjustment suggestions; perform a secondary quality assessment on the adjusted and optimized interframe animation, and perform corresponding processing based on the secondary quality assessment results.

[0027] Step S100 includes: S101. Obtaining the action data of the virtual character from the database, and the action data is based on motion capture technology or manually designed animation data sets, each data includes a time series of action data, wherein each time point corresponds to the posture information of each joint or bone of the character (such as the angle and position of the joint, etc.). The action data is composed of a number of frames, each frame represents the posture of the virtual character at a certain moment, and each action data is represented as an action sequence P, and P={p1,p2,...,pn}, wherein p1 represents the posture of the virtual character in the first frame, p2 represents the posture of the virtual character in the second frame, and so on, pn represents the posture of the virtual character in the nth frame, and n represents the number of frames in the action sequence P; S102. Analyze the acquired action data of the virtual character, and identify the key frames of each action of the virtual character, wherein the key frames refer to the action start frame and the action end frame of the virtual character; the key frame identification process is as follows: According to each action sequence of the virtual character, the difference Δpi between the virtual character postures corresponding to frame pi and frame pi+1 is calculated, and Δpi=||(pi+1)-pi||, where ||(pi+1)-pi|| represents the Euclidean distance between frame pi and frame pi+1, which is used to quantify the degree of change of the action of the virtual character between the two frames. If the action has changed significantly, the difference Δpi will be large, otherwise it will be small; summarize the differences Δpi, and draw the corresponding difference change curve Q1, and the horizontal axis of the difference change curve Q1 is the sequential number of the differences Δpi, and the vertical axis of the difference change curve Q1 is the difference Δpi; if a certain section of the difference change curve Q1 is the first If a certain segment of the difference change curve Q1 shows an increasing trend, the size relationship between the vertical axis Δpi of the corresponding difference change curve segment and the threshold ΔTstart is compared. If Δpi is greater than the threshold ΔTstart, the frame pi+1 is marked as the action start frame p_start of the virtual character; if a segment of the difference change curve Q1 shows a decreasing trend for the first time, the size relationship between the vertical axis Δpi of the corresponding difference change curve segment and the threshold ΔTend is compared. If Δpi is less than the threshold ΔTend, the frame pi+1 is marked as the action end frame p_end of the virtual character; for each action sequence of the virtual character, the corresponding action start frame and action end frame are summarized to identify the key frame of the corresponding action sequence; In this embodiment, since the action amplitude of the virtual character corresponding to each action sequence may be different, the difference thresholds ΔTstart and ΔTend at the start and end of the action may need to be adjusted; the appropriate thresholds are determined by calculating the difference distribution of the action sequence: Calculate the difference Δpi between each frame, and count the mean and standard deviation μ and σ of the difference Δpi; set the dynamic thresholds to ΔTstart = μ + kσ and ΔTend = μ − kσ, where k is a constant, which can usually be 1 or 2.

[0028] S103. According to the key frame of each action of the identified virtual character, the action data of the virtual character is divided to obtain the key frame data corresponding to the action data of the virtual character, and the key frame data corresponding to each action data of the virtual character are summarized to establish a key frame data set G of the virtual character, and G={g1,g2,...,gm}, wherein g1 represents the key frame data of the first action of the virtual character, g2 represents the key frame data of the second action of the virtual character, and so on, gm represents the key frame data of the mth action of the virtual character, and m represents the action number of the virtual character.

[0029] By analyzing the action data of the virtual character and identifying the key frames, the entire action sequence can be divided into a more streamlined key frame data set, thereby reducing the amount of calculations that need to be processed. By calculating the differences between action frames and using this as a basis for identifying key frames, redundant frames that have little impact on the action performance can be effectively eliminated, improving the performance quality of the virtual character's actions; in the rendering of animation effects, reducing unnecessary intermediate frames helps to reduce the unnatural and unsmooth transition effects. Key frame recognition provides an important reference for subsequent action generation and editing, especially in animation production and game development, where the animation of virtual characters usually relies on key frames; by accurately identifying the start and end frames of the action, it can help animators and developers more accurately control the smoothness and expressiveness of the character's actions.

[0030] Step S200 includes: S201. Obtaining a behavior instruction of a virtual character, wherein the behavior instruction of the virtual character is composed of an action number of the corresponding virtual character, and the behavior instruction of the virtual character is represented by B, and B={b1,b2,...,bk}, wherein b1 represents the first action number of the virtual character, b2 represents the second action number of the virtual character, and so on, bk represents the second action number of the virtual character; k represents the total number of action numbers corresponding to the behavior instruction of the virtual character, and k≤m; searching in the key frame data set G according to the behavior instruction of the virtual character, so as to obtain the key frame data sequence corresponding to the behavior instruction of the virtual character. Column, obtain corresponding action data in the database according to the key frame data sequence, mark the key frame position in the action data, filter the action data based on the key frame mark, retain the data from the action start frame to the action end frame for each action data, sort the filtered action data according to the behavior instructions of the virtual character, so as to obtain the motion sequence D of the virtual character corresponding to the behavior instructions, and D={d1,d2,...,dk}, and d1 represents the motion data corresponding to the behavior instruction b1, d2 represents the motion data corresponding to the behavior instruction b2, and so on, dk represents the motion data corresponding to the behavior instruction bk; S202. Perform physical simulation on the motion sequence D of the virtual character to obtain the physical properties of the virtual character at each key frame moment; for two adjacent actions bj and bj+1 in the motion sequence of the virtual character, obtain the action end frame pj_end of bj and the action start frame pj+1_start of bj+1 according to the corresponding key frame data, generate the transition frame pt between the adjacent actions bj and bj+1 by interpolation method, and the corresponding calculation formula is: p=(1-α)·pj_end+α·pj+1_start; wherein α represents the interpolation coefficient, and the value is [0,1]; according to the physical properties of the virtual character at each key frame moment, mark the motion data corresponding to the two adjacent actions bj and bj+1 in the motion sequence of the virtual character, mark the key frames and corresponding physical states of the adjacent actions bj and bj+1 and the generated transition frames, and use these data as training data; through the trained model, input the end frame and start frame of the two adjacent actions, the model will generate the corresponding interpolation frame sequence, combine the interpolation frame sequence with the motion sequence D, and obtain the interpolation frame animation between the two actions of the virtual character.

[0031] In this embodiment, the specific content of the physical simulation of the motion sequence D of the virtual character is: the motion sequence of the virtual character is physically simulated through a known physical engine (such as a model based on mass, force, and acceleration). The physical properties of each action will be calculated at each key frame moment, including speed, acceleration, gravity, etc. For example, the Newton-Euler method or rigid body dynamics model is used to calculate the physical state of the virtual character at each key frame moment, and the calculation model can be adjusted according to the body structure of the virtual character (such as joints, bones) and external forces (such as gravity, friction).

[0032] The generation process of the interpolation frame animation is as follows: Assume that the key frame data of the virtual character's actions bj and bj+1 include the following physical properties: Position p(t): the three-dimensional spatial coordinates of each joint of the virtual character's skeleton at time t; Velocity v(t): linear velocity of the virtual character's joints, in m / s;.

[0033] Acceleration a(t): Linear acceleration of the virtual character's joints, in m / s2; Rotation angle θ(t): the rotation angle of the character's joint, in degrees or radians; Angular velocity ω(t): The angular velocity of the character's joints, in rad / s.

[0034] Transition frames (such as position, velocity, acceleration, etc.) generated by physical simulation and interpolation are used to train the model. The input of each training sample includes the following: Input data: the end frame and start frame of two adjacent actions; Target data: the actual data of the transition frame, including the position information, speed, acceleration and other physical properties of the transition frame.

[0035] The end frames and start frames of adjacent actions can be used as training sample inputs, and the generated transition frame data can be used as target outputs.

[0036] By using machine learning (such as deep neural networks, convolutional neural networks, etc.) for training, the model learns how to generate a sequence of complementary frames based on the key frames of adjacent actions input. The input of the model is the start frame and end frame of the action (including position, speed, acceleration, etc.), and the output is the transition frame.

[0037] Step S300 includes: S301. Obtain the interpolation frame animation between two actions of the virtual character, obtain the corresponding motion data of the virtual character for each transition frame p corresponding to the interpolation frame animation, and calculate the motion change ΔMp(t) between two adjacent transition frames, where the specific formula is: ΔMp(t)={Σe∈[1,E],[Mep(t+1)-Mep(t)] 2}^(1 / 2); Where t represents the time point, Mep(t+1) represents the e-th kinematic feature of the transition frame corresponding to the time point t+1, and Mep(t) represents the e-th kinematic feature of the transition frame corresponding to the time point t; E represents the number of kinematic features, and the kinematic features include the joint or bone positions of the virtual character, the changes in joint angles, the rotation center of the skeleton, etc.; the motion changes ΔMp(t) between all adjacent transition frames are summarized, and the motion change curve Q2 is drawn with the time point as the horizontal axis and the motion change ΔMp(t) as the vertical axis, and the tangent slope R corresponding to each data point on the motion change curve Q2 is calculated, and the tangent slope R corresponding to each data point is compared with the set threshold Rth, and the transition frame of the data point whose tangent slope R is greater than the set threshold Rth is marked as the focus frame; S302. Summarize all the focus frames in the interpolation animation between two actions of the virtual character, and for each focus frame, calculate the smoothness index S of the focus frame in combination with the motion change ΔMp(t) corresponding to each transition frame p in the interpolation animation. The corresponding calculation formula is: S(pt)=1 / {1+Σt∈[1,T],ΔMp(t)·β}; Among them, T represents the total number of time points corresponding to the interframe animation between two actions of the virtual character, which is equal to the number of transition frames corresponding to the interframe animation between two actions of the virtual character; β represents the smoothness weight, and it ranges from 0 to 1; the physical consistency score SW corresponding to each focus frame is calculated, and the specific calculation formula is: SW(pt)=1-|θp(t+1)-θp(t)| / (θ_max-θ_min); Among them, θp(t+1) represents the joint angle value of the virtual character in the transition frame corresponding to the time point t+1, θp(t) represents the joint angle value of the virtual character in the transition frame corresponding to the time point t; θ_max-θ_min represents the maximum angle variation range allowed for the joints of the virtual character; the action consistency index C corresponding to the focus frame in the interpolation frame animation between two actions of the virtual character is calculated, and the corresponding calculation formula is: C(pt)={(1 / T)Σt∈[1,T],Σe∈[1,E],||Me(pt)-M(pt)||} / T; Where M(pt) represents the expected motion characteristics of the transition frame corresponding to time point t; S303. According to the smoothness index S, physical consistency score SW and motion consistency index C of the focus frame, calculate the corresponding comprehensive quality score SQ, and SQ(pt)=w1·S(pt)+w2·SW(pt)+w3·C(pt); wherein w1, w2 and w3 are the weight coefficients of the smoothness index S, the physical consistency score SW and the motion consistency index C respectively, and the sum of the weights is 1; compare the comprehensive quality score SQ of the focus frame with the threshold SQ0, and mark the focus frame whose comprehensive quality score SQ is less than the threshold SQ0 as a problem frame.

[0038] Step S400 includes: S401. Obtain the problem frame of the interpolation animation between two actions of the virtual character, extract the smoothness index S, physical consistency score SW and action consistency index C corresponding to the problem frame, and compare the smoothness index S, physical consistency score SW and action consistency index C with the corresponding smoothness index threshold S0, physical consistency score threshold SW0 and action consistency index threshold C0 respectively; if the smoothness index S is less than the smoothness index threshold S0, it means that the defect corresponding to the problem frame is that the transition speed is too fast or too slow, and the improvement suggestion is "optimize the interpolation algorithm or increase or decrease the number of interpolation frames"; if the physical consistency score SW is less than the physical consistency score threshold SW0, it means that the defect corresponding to the problem frame is a violation of the physical law, and the improvement suggestion is "adjust the physical properties of the interpolation animation"; if the action consistency index C is greater than the action consistency index threshold C0, it means that the defect corresponding to the problem frame is inconsistent behavior of the virtual character, and the improvement suggestion is "adjust the transition logic between actions, and optimize the selection and sorting of action sequences"; S402. Output the problem frames and the corresponding adjustment suggestions to relevant personnel, who adjust and optimize the interpolation frame animation based on the adjustment suggestions; and perform a secondary quality assessment on the adjusted interpolation frame animation, obtain the comprehensive quality score SQ1 corresponding to the secondary quality assessment, and compare the comprehensive quality score SQ1 with the threshold value SQ0; if the comprehensive quality score SQ1 is less than the threshold value SQ0, repeat the content of S401 until the corresponding comprehensive quality score is greater than or equal to the threshold value SQ0.

[0039] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0040] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for simulating virtual character behavior based on artificial intelligence, characterized in that: The method comprises the following steps: Step S100. Acquire action data of the virtual character from the database, analyze the acquired action data of the virtual character, thereby obtaining key frame data of each action of the virtual character, and establish a key frame data set of the virtual character based on the key frame data; Step S200. Obtaining the behavior instruction of the virtual character, searching in the key frame data set based on the behavior instruction of the virtual character, thereby obtaining the motion sequence of the virtual character corresponding to the behavior instruction; generating a transition frame between two adjacent actions in the motion sequence according to the motion sequence of the virtual character, thereby obtaining a frame-filling animation between two actions of the virtual character; Step S300. Obtain motion data corresponding to the interlaced frame animation between two actions of the virtual character, analyze the motion data corresponding to the interlaced frame animation, and thereby identify the focus frame in the interlaced frame animation; perform quality assessment on the focus frame in combination with the motion data; and identify the problem frame in the interlaced frame animation based on the quality assessment result; Step S400. Analyze the problem frames in the interframe animation, generate corresponding adjustment suggestions, and output the problem frames and the corresponding adjustment suggestions to relevant personnel, who then adjust and optimize the interframe animation based on the adjustment suggestions; perform a secondary quality assessment on the adjusted and optimized interframe animation, and perform corresponding processing based on the secondary quality assessment results.

2. The method for simulating virtual character behavior based on artificial intelligence according to claim 1, characterized in that: The step S100 includes: S101. Acquire action data of a virtual character from a database, wherein the action data is composed of a plurality of frames, each frame representing a posture of the virtual character at a certain moment, and each action data is represented as an action sequence P, and P={p1,p2,...,pn}, wherein p1 represents the posture of the virtual character in the first frame, p2 represents the posture of the virtual character in the second frame, and so on, pn represents the posture of the virtual character in the nth frame, and n represents the number of frames in the action sequence P; S102. Analyze the acquired action data of the virtual character, and identify the key frames of each action of the virtual character, wherein the key frames refer to the action start frame and the action end frame of the virtual character; the key frame identification process is as follows: According to each action sequence of the virtual character, the difference Δpi between the virtual character postures corresponding to frame pi and frame pi+1 is calculated, and Δpi=||(pi+1)-pi||, where ||(pi+1)-pi|| represents the Euclidean distance between frame pi and frame pi+1; the difference Δpi is summarized, and the corresponding difference change curve Q1 is drawn, and the horizontal axis of the difference change curve Q1 is the sequential number of the difference Δpi, and the vertical axis of the difference change curve Q1 is the difference Δpi; if a certain section of the difference change curve Q1 shows an increasing trend for the first time, compare the vertical axis Δpi of the corresponding difference change curve section with The size relationship between the threshold ΔTstart, if Δpi is greater than the threshold ΔTstart, the frame pi+1 is marked as the action start frame p_start of the virtual character; if a certain segment of the difference change curve Q1 shows a decreasing trend for the first time, the size relationship between the vertical axis Δpi of the corresponding difference change curve segment and the threshold ΔTend is compared, if Δpi is less than the threshold ΔTend, the frame pi+1 is marked as the action end frame p_end of the virtual character; for each action sequence of the virtual character, the corresponding action start frame and action end frame are summarized to identify the key frame of the corresponding action sequence; S103. According to the key frame of each action of the identified virtual character, the action data of the virtual character is divided to obtain the key frame data corresponding to the action data of the virtual character, and the key frame data corresponding to each action data of the virtual character are summarized to establish a key frame data set G of the virtual character, and G={g1,g2,...,gm}, wherein g1 represents the key frame data of the first action of the virtual character, g2 represents the key frame data of the second action of the virtual character, and so on, gm represents the key frame data of the mth action of the virtual character, and m represents the action number of the virtual character.

3. The method for simulating virtual character behavior based on artificial intelligence according to claim 2, characterized in that: The step S200 includes: S201. Obtaining a behavior instruction of a virtual character, wherein the behavior instruction of the virtual character is composed of an action number of the corresponding virtual character, and the behavior instruction of the virtual character is represented by B, and B={b1,b2,...,bk}, wherein b1 represents the first action number of the virtual character, b2 represents the second action number of the virtual character, and so on, bk represents the second action number of the virtual character; k represents the total number of action numbers corresponding to the behavior instruction of the virtual character, and k≤m; searching in a key frame data set G according to the behavior instruction of the virtual character, so as to obtain a key frame data sequence corresponding to the behavior instruction of the virtual character. Column, obtain corresponding action data in the database according to the key frame data sequence, mark the key frame position in the action data, filter the action data based on the key frame mark, retain the data from the action start frame to the action end frame for each action data, sort the filtered action data according to the behavior instructions of the virtual character, so as to obtain the motion sequence D of the virtual character corresponding to the behavior instructions, and D={d1,d2,...,dk}, and d1 represents the motion data corresponding to the behavior instruction b1, d2 represents the motion data corresponding to the behavior instruction b2, and so on, dk represents the motion data corresponding to the behavior instruction bk; S202. Perform physical simulation on the motion sequence D of the virtual character, so as to obtain the physical properties of the virtual character at each key frame moment; for two adjacent actions bj and bj+1 in the motion sequence of the virtual character, obtain the action end frame pj_end of bj and the action start frame pj+1_start of bj+1 according to the corresponding key frame data, and generate the transition frame pt between the adjacent actions bj and bj+1 by interpolation method, and the corresponding calculation formula is: p=(1-α)·pj_end+α·pj+1_start; wherein α represents the interpolation coefficient, and the value is [0,1]; according to the physical properties of the virtual character at each key frame moment, mark the motion data corresponding to the two adjacent actions bj and bj+1 in the motion sequence of the virtual character, mark the key frames and corresponding physical states of the adjacent actions bj and bj+1 and the generated transition frames, and use these data as training data; through the trained model, input the end frame and start frame of the two adjacent actions, the model will generate the corresponding interpolation frame sequence, combine the interpolation frame sequence with the motion sequence D, so as to obtain the interpolation frame animation between the two actions of the virtual character.

4. The method for simulating virtual character behavior based on artificial intelligence according to claim 3, characterized in that: The step S300 includes: S301. Obtain the interpolation frame animation between two actions of the virtual character, obtain the corresponding motion data of the virtual character for each transition frame p corresponding to the interpolation frame animation, and calculate the motion change ΔMp(t) between two adjacent transition frames, where the specific formula is: ΔMp(t)={Σe∈[1,E],[Mep(t+1)-Mep(t)] 2 }^(1 / 2); Where t represents the time point, Mep(t+1) represents the e-th kinematic feature of the transition frame corresponding to the time point t+1, and Mep(t) represents the e-th kinematic feature of the transition frame corresponding to the time point t; E represents the number of kinematic features; the motion changes ΔMp(t) between all adjacent transition frames are summarized, and the motion change curve Q2 is drawn with the time point as the horizontal axis and the motion change ΔMp(t) as the vertical axis, and the tangent slope R corresponding to each data point on the motion change curve Q2 is calculated, and the tangent slope R corresponding to each data point is compared with the set threshold Rth, and the transition frame of the data point whose tangent slope R is greater than the set threshold Rth is marked as the focus frame; S302. Summarize all the focus frames in the interpolation animation between two actions of the virtual character, and for each focus frame, calculate the smoothness index S of the focus frame in combination with the motion change ΔMp(t) corresponding to each transition frame p in the interpolation animation. The corresponding calculation formula is: S(pt)=1 / {1+Σt∈[1,T],ΔMp(t)·β}; Among them, T represents the total number of time points corresponding to the interframe animation between two actions of the virtual character, which is equal to the number of transition frames corresponding to the interframe animation between two actions of the virtual character; β represents the smoothness weight, and it ranges from 0 to 1; the physical consistency score SW corresponding to each focus frame is calculated, and the specific calculation formula is: SW(pt)=1-|θp(t+1)-θp(t)| / (θ_max-θ_min); Among them, θp(t+1) represents the joint angle value of the virtual character in the transition frame corresponding to the time point t+1, θp(t) represents the joint angle value of the virtual character in the transition frame corresponding to the time point t; θ_max-θ_min represents the maximum angle variation range allowed for the joints of the virtual character; the action consistency index C corresponding to the focus frame in the interpolation frame animation between two actions of the virtual character is calculated, and the corresponding calculation formula is: C(pt)={(1 / T)Σt∈[1,T],Σe∈[1,E],||Me(pt)-M(pt)||} / T; Where M(pt) represents the expected motion characteristics of the transition frame corresponding to time point t; S303. According to the smoothness index S, physical consistency score SW and motion consistency index C of the focus frame, calculate the corresponding comprehensive quality score SQ, and SQ(pt)=w1·S(pt)+w2·SW(pt)+w3·C(pt); wherein w1, w2 and w3 are the weight coefficients of the smoothness index S, the physical consistency score SW and the motion consistency index C respectively, and the sum of the weights is 1; compare the comprehensive quality score SQ of the focus frame with the threshold SQ0, and mark the focus frame whose comprehensive quality score SQ is less than the threshold SQ0 as a problem frame.

5. The method for simulating virtual character behavior based on artificial intelligence according to claim 4, characterized in that: The step S400 includes: S401. Obtain the problem frame of the interpolation animation between two actions of the virtual character, extract the smoothness index S, physical consistency score SW and action consistency index C corresponding to the problem frame, and compare the smoothness index S, physical consistency score SW and action consistency index C with the corresponding smoothness index threshold S0, physical consistency score threshold SW0 and action consistency index threshold C0 respectively; if the smoothness index S is less than the smoothness index threshold S0, it means that the defect corresponding to the problem frame is that the transition speed is too fast or too slow, and the improvement suggestion is "optimize the interpolation algorithm or increase or decrease the number of interpolation frames"; if the physical consistency score SW is less than the physical consistency score threshold SW0, it means that the defect corresponding to the problem frame is a violation of the physical law, and the improvement suggestion is "adjust the physical properties of the interpolation animation"; if the action consistency index C is greater than the action consistency index threshold C0, it means that the defect corresponding to the problem frame is inconsistent behavior of the virtual character, and the improvement suggestion is "adjust the transition logic between actions, and optimize the selection and sorting of action sequences"; S302. Output the problem frames and the corresponding adjustment suggestions to relevant personnel, who will adjust and optimize the interpolated frame animation based on the adjustment suggestions; and perform a secondary quality assessment on the adjusted interpolated frame animation to obtain a comprehensive quality score SQ1 corresponding to the secondary quality assessment, and compare the comprehensive quality score SQ1 with the threshold SQ0; if the comprehensive quality score SQ1 is less than the threshold SQ0, repeat the content of S401 until the corresponding comprehensive quality score is greater than or equal to the threshold SQ0.

6. A virtual character behavior and action simulation system based on artificial intelligence, applied to a virtual character behavior and action simulation method based on artificial intelligence as claimed in any one of claims 1 to 5, characterized in that: The system comprises: an action data acquisition and analysis module, a behavior instruction parsing and motion sequence generation module, a frame-filling animation generation module, a frame-filling animation quality assessment and problem identification module, and a frame-filling animation optimization and adjustment module; The action data acquisition and analysis module acquires the action data of the virtual character from the database, analyzes the acquired action data of the virtual character, thereby obtaining key frame data of each action of the virtual character, and establishes a key frame data set of the virtual character according to the key frame data; The behavior instruction parsing and motion sequence generating module obtains the behavior instruction of the virtual character, searches in the key frame data set based on the behavior instruction of the virtual character, and thus obtains the motion sequence of the virtual character corresponding to the behavior instruction; The interpolation frame animation generation module generates a transition frame between two adjacent actions in the motion sequence according to the motion sequence of the virtual character, thereby obtaining an interpolation frame animation between two actions of the virtual character; The interframe animation quality assessment and problem identification module obtains motion data corresponding to the interframe animation between two actions of the virtual character, analyzes the motion data corresponding to the interframe animation, and thus identifies the focus frame in the interframe animation; performs quality assessment on the focus frame in combination with the motion data; and identifies the problem frame in the interframe animation based on the quality assessment result; The frame-interpolation animation optimization and adjustment module analyzes the problem frames in the frame-interpolation animation, generates corresponding adjustment suggestions, and outputs the problem frames and the corresponding adjustment suggestions to relevant personnel, who then adjust and optimize the frame-interpolation animation based on the adjustment suggestions; performs a secondary quality assessment on the adjusted and optimized frame-interpolation animation, and performs corresponding processing based on the secondary quality assessment results.

7. The artificial intelligence-based virtual character behavior simulation system according to claim 6, characterized in that: The motion data acquisition and analysis module includes a motion data acquisition unit, a motion data analysis unit and a key frame data set construction unit; The action data acquisition unit acquires the action data of the virtual character from a database; the action data analysis unit analyzes the acquired action data, calculates the difference between action frames, draws a difference change curve, and identifies the key frame of each action according to a threshold; the key frame data set construction unit establishes a key frame data set of the virtual character based on the key frame data.

8. The artificial intelligence-based virtual character behavior and action simulation system according to claim 6, characterized in that: The behavior instruction parsing and motion sequence generating module includes a behavior instruction acquiring unit and a motion sequence generating unit; The behavior instruction acquisition unit receives the behavior instruction of the virtual character; the motion sequence generation unit searches and acquires the corresponding key frame data sequence in the key frame data set according to the behavior instruction, and generates the motion sequence of the virtual character by screening and sorting; The interpolation frame animation generation module includes a physical simulation unit, a transition frame generation unit and an interpolation frame animation generation unit; The physical simulation unit simulates the physical properties of the virtual character at each key frame moment, thereby obtaining the physical properties of the virtual character at each key frame moment; the transition frame generation unit generates a transition frame between two actions based on the motion sequence of the virtual character and the physical simulation; The interpolation frame animation generation unit collects transition frames between two adjacent actions of the virtual character, thereby generating an interpolation frame animation.

9. The virtual character behavior and action simulation system based on artificial intelligence according to claim 6, characterized in that: The frame-interpolation animation quality assessment and problem identification module includes a motion variation calculation unit, a focus frame identification unit, and a problem frame marking unit; The motion change calculation unit calculates the motion change between each transition frame in the interpolated frame animation; the focus frame identification unit identifies the focus frame according to the motion change; the problem frame labeling unit calculates the comprehensive quality score of the focus frame, identifies and labels the problem frame. The focus frame identification unit identifies the focus frame according to the motion change; the problem frame labeling unit calculates the comprehensive quality score of the focus frame, identifies the problem frame according to the comprehensive quality score, and labels the problem frame.

10. The artificial intelligence-based virtual character behavior simulation system according to claim 6, characterized in that: The frame-filling animation optimization and adjustment module includes a problem frame analysis unit, an optimization suggestion generation unit and a secondary quality assessment unit; The problem frame analysis unit analyzes the motion data of the problem frame and identifies the corresponding defects; the optimization suggestion generation unit generates corresponding adjustment suggestions to relevant personnel based on the analysis results of the problem frame analysis unit, and the relevant personnel make corresponding optimization adjustments; the secondary quality assessment unit performs a secondary quality assessment on the adjusted frame-filling animation and performs corresponding processing based on the secondary quality assessment results.

Images