Method for intelligently controlling digital human through action label

Abstract

The invention discloses a method for intelligently controlling a digital human through an action tag, and relates to the technical field of intelligent control. The method comprises the following steps: S1, determining the range and purpose of actions, classifying the actions, and determining which specific application scenes and target user groups are to be used by an action library; s2, designing a technical label system used for describing various action features and providing a basis for technical action recognition and analysis; s3, establishing a behavior control model of the digital character according to the tagged action description; and S4, through cross validation, dividing the data set into a plurality of subsets for multiple times of training and verification, and adjusting the structure of the model according to an evaluation result. According to the method, the naturalness and fluency of digital character expression can be improved, the flexibility and customization capability of the technology are enhanced, and intelligent action selection and transition are realized, so that better user experience and technical support are provided for various application scenes.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to a method for intelligently controlling a digital human using action tags. Background Art

[0002] Motion capture technology is the key foundation for designing motion drivers for digital humans. It uses sensors and cameras to record movements in the real world and converts them into digital formats. This technology has been widely used in film and television production, game development, and virtual reality. It can accurately capture and reproduce the movements of humans and animals, providing high-quality basic materials for the construction of motion libraries. Although the captured movements can be improved through post-editing and adjustment, this usually requires professional skills and additional time investment. The processed movements may still need to be manually corrected or fine-tuned to suit the needs of specific application scenarios or roles, increasing the development and production cycle.

[0003] Secondly, motion synthesis and control technology are crucial to achieving complex motion sequences. Motion synthesis uses algorithms and models to generate new motion sequences, which can be created based on existing motion clips or parametric descriptions. At the same time, motion control technology ensures that digital characters can show natural movements and reactions in different environments and interactive situations through real-time feedback technology and dynamic motion planning, thereby enhancing user experience and interactivity. For applications that require real-time generation or interactive control, such as virtual reality and real-time games, motion synthesis technology must be able to respond quickly to user input or environmental changes. Traditional motion synthesis technology may not meet these requirements because its generation process may require long computing time or preprocessing steps. Motion synthesis and control technology often needs to be integrated with other technologies (such as physics engines, artificial intelligence, graphics rendering, etc.) to achieve more advanced functions and effects. This integration process may lead to increased technical complexity and put forward higher technical requirements and challenges for developers and production teams.

[0004] After investigating the existing technologies, it is found that the existing digital character behavior control technologies have the following main shortcomings:

[0005] Limitations and subjectivity of labeled action libraries: Existing methods often rely on manually constructed labeled action libraries. These libraries may be limited by the developer's subjective choices and limited coverage, resulting in digital characters that behave naturally in certain scenarios but may appear stiff or inappropriate in other scenarios.

[0006] Unnaturalness in action selection and combination: In the process of action selection and combination, existing technologies may not be able to achieve a natural transition of actions, or lack of logic between different actions. This causes digital characters to appear clumsy or incoherent when expressing complex emotions or complex actions.

[0007] Coping with challenges in new scenarios and situations: Existing technologies are difficult to flexibly cope with new application scenarios or complex interactive situations. For example, when digital characters need to react according to real-time situations, existing technologies may not be able to achieve instant and accurate action adjustments.

[0008] Therefore, a method of intelligently controlling digital human with action tags is proposed. Summary of the invention

[0009] The purpose of the present invention is to solve the problems mentioned in the above background technology, and provide a method for intelligently controlling a digital human using action tags.

[0010] In order to achieve the above-mentioned purpose, the present invention specifically adopts the following technical solution: a method for intelligently controlling a digital human using an action tag, comprising the following steps:

[0011] Step S1: Determine the scope and purpose of the action and classify the actions, and clarify the specific application scenarios and target user groups that the action library will be used for;

[0012] Step S2: Design a technical label system to describe various action features and provide a basis for technical action recognition and analysis;

[0013] Step S3: Building a behavior control model of the digital character based on the labeled action description;

[0014] Step S4: Through cross-validation, the data set is divided into multiple subsets for multiple training and validation. According to the evaluation results, the structure of the model is adjusted, the number of layers is increased or decreased, the number of nodes in each layer is adjusted, or the network connection method is changed to better adapt to the data characteristics and the complexity of the problem;

[0015] Step S5: Intelligent connection and transition of actions: Seamlessly connect the output single actions together to make them look smooth and coherent visually and dynamically without any abrupt or unnatural feeling, and make the action transition smoother by adjusting the position and interpolation method between key frames.

[0016] Step S6: Unified drive of digital human behavior: It can select appropriate actions according to the current scene, task or user input; it can also dynamically adjust the behavior of the digital character according to the user's input or command, including responding to the user's control instructions, interactive selection or real-time user feedback to adjust the character's action performance.

[0017] Furthermore, the action library in step S1 includes the following categories:

[0018] 1. Basic movements:

[0019] Movement: walking, running, jumping, crawling, sliding, turning;

[0020] Posture movements: sitting, standing, lying, bending, leaning over, and turning sideways;

[0021] 2. Compound Actions:

[0022] Operation actions: open door, close door, open window, close window, take, place, carry;

[0023] Interactions: handshake, hug, wave, fist, touch, hug;

[0024] 3. Expressions and emotional movements:

[0025] Facial expressions: smiling, angry, surprised, sad, confused;

[0026] Posture expression: confident, anxious, relaxed, nervous, focused;

[0027] 4. Social and daily life activities:

[0028] Social actions: greeting, saying goodbye, talking, winking;

[0029] Daily life activities: eating, drinking, washing, working, and resting;

[0030] 5. Professional field actions:

[0031] Medical field: examination, treatment, and surgical operation;

[0032] Industrial field: assembly, maintenance and operation of machines;

[0033] Artistic performances: dance movements, drama performances, music performances;

[0034] 6. Environmental Interaction Actions:

[0035] Indoor space: entering a room, leaving a room, approaching an object, moving away from an object;

[0036] Outdoor spaces: through traffic, through people, through natural obstacles;

[0037] 7. Action attributes:

[0038] Speed ​​and intensity: fast, slow, gentle, strong;

[0039] Precision and accuracy: precise, imprecise, high or low accuracy.

[0040] Furthermore, the technically designed label system includes the following steps:

[0041] Step S2.1: Define the attributes and classification of the tags. First, consider the type of action, such as static or dynamic action. Second, determine the direction of the action, including forward, backward, left, right, and up and down. Finally, mark the speed of the action, from fast to slow.

[0042] Step S2.2: determining the structure and format of the label, determining the structure and format of the label to describe each action in detail by the type of action (walking, running, jumping, turning), direction (including forward, backward, left, right, up, down, clockwise, counterclockwise), and speed (fast, medium, slow levels);

[0043] Step S2.3: Design the naming convention of the label;

[0044] Step S2.4: Consider action combinations and composite labels.

[0045] Furthermore, step S3 specifically includes the following steps: supervised learning learns the association between action label sequences from labeled data and is trained with carefully labeled data sets to learn the significant relationship between input data and predefined action labels; the algorithm builds a model by analyzing a large number of labeled examples, which contain multiple variants and scenarios of different actions; by learning these data, the supervised learning algorithm can infer the patterns and regularities of different action features; during the process of model training, the algorithm will continuously optimize its internal parameters to accurately predict the action label sequence in new data to the greatest extent possible.

[0046] Furthermore, the establishment of the behavior control model also includes feature extraction, and the feature extraction includes the following steps:

[0047] 1. Data preprocessing:

[0048] Data cleaning: remove outliers and process missing data;

[0049] Data smoothing: Use filters or smoothing techniques to smooth data and remove high-frequency noise;

[0050] Standardization: ensure that the data is on the same scale to avoid the impact of different feature value ranges on the model;

[0051] 2. Time series feature extraction:

[0052] If the data is time series, i.e. action sequence or motion capture data, use the following time series features:

[0053] Statistical characteristics: such as mean, variance, maximum, minimum, standard deviation, etc. These characteristics can describe the distribution and central tendency of data;

[0054] Time domain features: including the acceleration, angle, speed, etc. of the action. These features can be calculated directly from the original data;

[0055] Frequency domain features: Frequency domain features are extracted through Fourier transform and other techniques to analyze the periodicity or frequency distribution of actions;

[0056] 3. Spatial feature extraction:

[0057] For tasks involving spatial position or posture, such as human action recognition, the following spatial features are used:

[0058] Joint angle: Calculate the angle from the joint position data;

[0059] Joint velocity and acceleration: Calculate the joint velocity and acceleration by the rate of change of joint position;

[0060] Pose descriptor: a representation of human posture, based on keypoint-based posture descriptors or image features;

[0061] 4. Feature selection and dimensionality reduction:

[0062] Feature selection: Select the most relevant or representative features, automatically based on the model.

[0063] Dimensionality reduction: Use principal component analysis (PCA) to reduce the high-dimensional feature space to a lower dimension, retaining the most important information.

[0064] Furthermore, in step S4, the effect and performance of the model are measured by the following six evaluation indicators:

[0065] 1. Real-time responsiveness: measures the model's ability to respond immediately to user input, including processing speed and real-time performance;

[0066] 2. Semantic understanding accuracy: evaluates the model’s ability to understand user intent, including identifying the topic, sentiment, and specific requirements of the question;

[0067] 3. Information accuracy: Check whether the information provided by the model is accurate, complete and useful, including the accuracy of facts and the relevance of content;

[0068] 4. Action generation quality: evaluate the fluency, grammatical correctness and clarity of the actions generated by the model;

[0069] 5. Behavioral consistency: Check whether the model can maintain consistency in behavior and context over a long period of time;

[0070] 6. User satisfaction: Measure the user's overall satisfaction with the model interaction experience through user feedback or questionnaires.

[0071] Furthermore, the specific steps of step S5 are as follows:

[0072] Step S5.1: Analyze the action sequence:

[0073] First, the action sequence to be connected needs to be analyzed;

[0074] Step S5.2: Adjust keyframes:

[0075] Determine the key frames for each action, including the start frame and the end frame;

[0076] Step S5.3: Interpolation and curve adjustment:

[0077] Use appropriate interpolation methods (such as linear interpolation, Bezier curves, etc.) to ensure smooth transitions between keyframes;

[0078] Step S5.4: Add transition frames:

[0079] Add transition frames between different actions, which include transition poses and transition actions.

[0080] Step S5.5: Time interval adjustment:

[0081] Adjust the time interval between keyframes as needed;

[0082] Step S5.6: Overall rhythm and logical consistency:

[0083] Make sure the duration of each action is consistent with the rhythm and logic of the overall animation or game;

[0084] Step S5.7: Flexible response to scenario requirements:

[0085] Depending on the needs of the specific scene, you may need to adjust the duration of the action to make the transition between different scenes more natural and seamless.

[0086] Furthermore, the step S6 selects an appropriate action according to the current scene, task or user input, including the following specific steps:

[0087] Step S6.1: Scene recognition and state judgment:

[0088] Able to accurately identify the current scene and role status of the digital character;

[0089] Step S6.2: Action selection and sequence creation:

[0090] According to the recognized character state, select appropriate basic actions from the predefined action library;

[0091] Step S6.3: Environmental Adaptability:

[0092] When environmental conditions change, the digital character's movements can be dynamically adjusted;

[0093] Step S6.4: User input and real-time feedback response:

[0094] Ability to respond to user control commands or interactive selections to adjust the digital character's behavior; this includes real-time user feedback, requiring the ability to instantly update the character's movements to meet user needs and operations.

[0095] Step S6.5: Action optimization and performance details:

[0096] Improve the realism and smoothness of actions through optimization algorithms.

[0097] The beneficial effects of the present invention are as follows:

[0098] 1. Natural and smooth action performance: By building a label action library and selecting, combining and controlling the transition of actions, it is possible to ensure that digital characters perform naturally and smoothly in different situations. This naturalness and smoothness are crucial for applications such as virtual reality (VR), augmented reality (AR), video games, and film and television special effects, and can enhance the user's sense of immersion and experience quality.

[0099] 2. High customizability and flexibility: It allows developers to customize the action library and behavior control according to specific needs, thus adapting to different types of digital characters and application scenarios. This flexibility allows the technology to be applied in a variety of application fields, including education, training simulation, entertainment, etc.

[0100] 3. Intelligent action selection and transition: Appropriate actions can be selected by reasoning or predicting user intent, and the transition between actions can be smooth and natural. For example, when a digital character needs to go from walking to sitting, the technology can intelligently select the appropriate transition action to avoid unnatural jumps or freezes.

[0101] 4. Real-time and responsiveness: In response to real-time interaction and application requirements, the technology can quickly respond to user input or environmental changes and adjust the behavior and actions of digital characters. This is especially important for applications that require dynamic interaction, such as real-time games or virtual meetings.

[0102] 5. Optimize resource utilization: Technical design can optimize resource utilization, such as the efficiency of processor and memory usage, to ensure stable performance and good user experience on various platforms (such as mobile devices, PCs, etc.). BRIEF DESCRIPTION OF THE DRAWINGS

[0103] Figure 1 It is a schematic flow chart of the method of the present invention;

[0104] Figure 2 It is a schematic diagram of the classification of the action library of the present invention;

[0105] Figure 3 It is a schematic diagram of the technical label system structure of the present invention;

[0106] Figure 4 It is a schematic diagram of the behavior control model of the present invention;

[0107] Figure 5 It is a schematic diagram of unified driving of the behavior of the present invention. DETAILED DESCRIPTION

[0108] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0109] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. 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.

[0110] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. In addition, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0111] The electrical components mentioned in this article are all connected to an external main controller and 220V mains electricity, and the main controller can be a conventional known device for controlling a computer or the like.

[0112] In the description of the embodiments of the present invention, it should be noted that the terms "inside", "outside", "upper", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, or are the orientations or positional relationships in which the inventive product is usually placed when used. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0113] like Figures 1 to 5 As shown, a method for intelligently controlling a digital human with action tags comprises the following steps:

[0114] Step S1: Determine the scope and purpose of the actions and classify the actions, and clarify the specific application scenarios and target user groups that the action library will be used for; Determining the scope and purpose of the action library is a key starting point for the design of the action library. This involves clarifying the specific application scenarios and target user groups that the action library will be used for. For example, the action library can be used in virtual reality games for character actions that enhance the player experience, or in educational applications to help learners imitate and understand complex physiological and mechanical actions. In addition, the action library can also support human-computer interaction technology for robot operation or motion feedback of virtual assistants, as well as in film and television production for animators and special effects artists to create realistic character actions and special effects. Therefore, determining the application scenarios and target user groups of the action library is crucial to ensuring the quality and practicality of the action library content to meet the needs and creative requirements of different fields.

[0115] Step S2: Design a technical label system to describe various action features and provide a basis for technical action recognition and analysis;

[0116] Step S3: Building a behavior control model of the digital character based on the labeled action description;

[0117] Step S4: Through cross-validation, the data set is divided into multiple subsets for multiple training and validation. According to the evaluation results, the structure of the model is adjusted, the number of layers is increased or decreased, the number of nodes in each layer is adjusted, or the network connection method is changed to better adapt to the data characteristics and the complexity of the problem;

[0118] Step S5: Intelligent connection and transition of actions: Seamlessly connect the output individual actions together so that they appear smooth and coherent visually and dynamically without any abrupt or unnatural feeling, and make the action transition smoother by adjusting the position and interpolation method between key frames. Avoid abrupt or unnatural feelings. Action transitions can be made smoother by adjusting the position and interpolation method between key frames. For example, by adding additional key frames to smooth speed changes, or adjusting the time interval between key frames to ensure a natural transition of actions. By adding appropriate transition frames, you can create a smoother transition effect between different actions. These frames can include transition poses or transition actions to ease the transition between two actions. Make sure that the duration of each action is consistent with the rhythm and logic of the overall animation or game. Depending on the needs of the scene, it may be necessary to shorten or lengthen the duration of the action to make the transition more natural.

[0119] Step S6: Digital human behavior unified drive: Ability to select appropriate actions based on the current scene, task or user input; ability to dynamically adjust the behavior of the digital character based on user input or commands, including responding to user control instructions, interactive selections or real-time user feedback to adjust the character's action performance. For example, in a game, based on the state of the character (such as attacking, defending, moving, etc.), the technology can automatically select appropriate actions to reflect these states. Different basic actions can be combined to create complex action sequences. For example, the transition from walking to running, or the smooth transition from standing to jumping. The action performance of the digital character is adjusted according to changes in the environment. For example, when the character walks from indoors to outdoors, the technology can automatically adjust the gait and speed of the walking action to reflect different terrain and lighting conditions.

[0120] The action library in step S1 includes the following categories:

[0121] 2. Basic actions:

[0122] Movement: walking, running, jumping, crawling, sliding, turning;

[0123] Posture movements: sitting, standing, lying, bending, leaning over, and turning sideways;

[0124] 2. Compound Actions:

[0125] Operation actions: open door, close door, open window, close window, take, place, carry;

[0126] Interactions: handshake, hug, wave, fist, touch, hug;

[0127] 3. Expressions and emotional movements:

[0128] Facial expressions: smiling, angry, surprised, sad, confused;

[0129] Posture expression: confident, anxious, relaxed, nervous, focused;

[0130] 4. Social and daily life activities:

[0131] Social actions: greeting, saying goodbye, talking, winking;

[0132] Daily life activities: eating, drinking, washing, working, and resting;

[0133] 5. Professional field actions:

[0134] Medical field: examination, treatment, and surgical operation;

[0135] Industrial field: assembly, maintenance and operation of machines;

[0136] Artistic performances: dance movements, drama performances, music performances;

[0137] 6. Environmental Interaction Actions:

[0138] Indoor space: entering a room, leaving a room, approaching an object, moving away from an object;

[0139] Outdoor spaces: through traffic, through people, through natural obstacles;

[0140] 7. Action attributes:

[0141] Speed ​​and intensity: fast, slow, gentle, strong;

[0142] Precision and accuracy: precise, imprecise, high or low accuracy.

[0143] Designing a technical labeling system includes the following steps:

[0144] Step S2.1: Define the attributes and classification of the tags. First, consider the type of action, such as static or dynamic action. Second, determine the direction of the action, including forward, backward, left, right, and up and down. Finally, mark the speed of the action, from fast to slow.

[0145] Step S2.2: Determine the structure and format of the label. Determine the structure and format of the label so that each action can be described in detail by the type of action (walking, running, jumping, turning), direction (including forward, backward, left, right, up, down, clockwise, counterclockwise), and speed (fast, medium, slow levels); this structured labeling technology provides a clear framework for accurate identification and analysis of actions, and is suitable for a variety of application scenarios, such as motion capture, virtual character control, and interaction design.

[0146] Step S2.3: Design a naming convention for labels; the naming convention for designing labels needs to ensure clarity, simplicity, and ease of understanding and application. Each label consists of action type, direction, and speed. For example, "Walk_Forward_Fast" indicates that the walking action is performed while moving forward at a fast speed, "Run_Left_Medium" indicates that the left-hand running action is performed at a medium speed, "Jump_Up_Fast" indicates that the upward jump is performed quickly, and "Turn_Clockwise_Medium" describes turning clockwise at a medium speed. This standardized labeling technology helps to accurately identify and apply different actions in various applications, and improves the operability and effectiveness of the technology.

[0147] Step S2.4: Consider action combinations and compound labels. When considering action combinations and compound labels, it is necessary to deal with the description requirements of complex or variable actions. Compound actions such as "walk_forward_medium speed+turn_clockwise_fast" combine walking and fast clockwise turning, which can accurately express the complexity of actions and simultaneous actions. Combined actions such as "walk_forward_fast, then jump_up" indicate the order and coherence of a series of actions, which helps to accurately simulate or capture multi-step action sequences. This flexible label design enables the technology to effectively identify and apply action combinations of different complexities to meet the needs of action recognition and control in various applications.

[0148] Step S3 specifically includes the following steps: supervised learning learns the association of action label sequences from labeled data and is trained with carefully labeled data sets to learn the significant relationship between input data and predefined action labels; the algorithm builds a model by analyzing a large number of labeled examples, which contain multiple variants and scenarios of different actions; by learning these data, the supervised learning algorithm can infer the patterns and regularities of different action features, such as the type of action, execution method, speed and direction, etc.; during the process of model training, the algorithm will continuously optimize its internal parameters to accurately predict the action label sequence in the new data to the greatest extent possible.

[0149] Establishing a behavior control model also includes feature extraction, which includes the following steps:

[0150] 2. Data preprocessing: Before feature extraction, the raw data needs to be preprocessed to ensure data quality and consistency.

[0151] Data cleaning: remove outliers and process missing data;

[0152] Data smoothing: Use filters or smoothing techniques to smooth data and remove high-frequency noise;

[0153] Standardization: ensure that the data is on the same scale to avoid the impact of different feature value ranges on the model;

[0154] 2. Time Series Feature Extraction

[0155] If the data is time series, i.e. action sequence or motion capture data, use the following time series features:

[0156] Statistical characteristics: such as mean, variance, maximum, minimum, standard deviation, etc. These characteristics can describe the distribution and central tendency of data;

[0157] Time domain features: including the acceleration, angle, speed, etc. of the action. These features can be calculated directly from the original data;

[0158] Frequency domain features: Frequency domain features are extracted through Fourier transform and other techniques to analyze the periodicity or frequency distribution of actions;

[0159] 3. Spatial Feature Extraction

[0160] For tasks involving spatial position or posture, such as human action recognition, the following spatial features are used:

[0161] Joint angle: Calculate the angle from the joint position data;

[0162] Joint velocity and acceleration: Calculate the joint velocity and acceleration by the rate of change of joint position;

[0163] Pose descriptor: a representation of human posture, based on keypoint-based posture descriptors or image features;

[0164] IV. Feature Selection and Dimensionality Reduction

[0165] After extracting a large number of features, feature selection or dimensionality reduction is required to reduce the computational complexity of the model and improve generalization ability:

[0166] Feature selection: Select the most relevant or representative features, automatically based on the model.

[0167] Dimensionality reduction: Use principal component analysis (PCA) to reduce the high-dimensional feature space to a lower dimension, retaining the most important information.

[0168] In step S4, the following six evaluation indicators are used to measure the effect and performance of the model:

[0169] 7. Real-time responsiveness: measures the model's ability to respond immediately to user input, including processing speed and real-time performance;

[0170] 8. Semantic understanding accuracy: evaluates the model’s ability to understand user intent, including identifying the topic, sentiment, and specific requirements of the question;

[0171] IX. Information accuracy: Check whether the information provided by the model is accurate, complete and useful, including the accuracy of facts and the relevance of content;

[0172] 10. Action generation quality: evaluate the fluency, grammatical correctness, and clarity of the actions generated by the model;

[0173] 11. Behavioral consistency: Check whether the model can maintain consistency in behavior and context over a long period of time;

[0174] User satisfaction: Measure the user's overall satisfaction with the model interaction experience through user feedback or questionnaires.

[0175] The specific steps of step S5 are as follows:

[0176] Step S5.1: Analyze the action sequence:

[0177] First, you need to analyze the action sequence to be connected; it is very important to understand the key frames, duration of each action, and the logical relationship between the actions;

[0178] Step S5.2: Adjust keyframes:

[0179] Determine the keyframes for each action, including the start and end frames; smooth the transition by adjusting the position and interpolation between keyframes. Consider adding additional in-between frames between keyframes to smooth out changes in speed and position.

[0180] Step S5.3: Interpolation and curve adjustment:

[0181] Use appropriate interpolation methods (such as linear interpolation, Bezier curves, etc.) to ensure smooth transitions between keyframes; the shape and slope of the curve will affect the fluidity and naturalness of the movement.

[0182] Step S5.4: Add transition frames:

[0183] Add transition frames between different actions. These frames include transition poses and transition actions. Transition frames help ease the transition between two actions, making the transition from one action to another smoother and more natural.

[0184] Step S5.5: Time interval adjustment:

[0185] Adjust the time between keyframes as needed. Sometimes increasing or decreasing the time between frames can make the transition smoother, especially in situations where fast or slow motion is desired.

[0186] Step S5.6: Overall rhythm and logical consistency:

[0187] Make sure the duration of each action matches the rhythm and logic of the overall animation or game; actions that are too long or too short can cause the animation to look unnatural or inconsistent.

[0188] Step S5.7: Flexible response to scenario requirements:

[0189] Depending on the needs of the specific scene, you may need to adjust the duration of the action to make the transition between different scenes more natural and seamless.

[0190] In step S6, an appropriate action is selected according to the current scene, task or user input, including the following specific steps:

[0191] Step S6.1: Scene recognition and state judgment:

[0192] Ability to accurately identify the current scene and character status of the digital character; for example, in a game, the technology needs to understand whether the character is attacking, defending, or moving.

[0193] Step S6.2: Action selection and sequence creation:

[0194] According to the recognized character state, appropriate basic actions are selected from the predefined action library; these basic actions can be walking, running, attacking, defending, jumping, etc. Complex action sequences can be created by combining different basic actions, such as the transition from walking to running, or a smooth transition from standing to jumping.

[0195] Step S6.3: Environmental Adaptability:

[0196] When environmental conditions change, the movement performance of the digital character can be dynamically adjusted; for example, when the character walks from indoors to outdoors, the gait and speed of the walking action can be automatically adjusted according to the terrain and lighting conditions, making it more realistic and practical.

[0197] Step S6.4: User input and real-time feedback response:

[0198] Ability to respond to user control commands or interactive selections to adjust the digital character's behavior; this includes real-time user feedback, requiring the ability to instantly update the character's movements to meet user needs and operations.

[0199] Step S6.5: Action optimization and performance details:

[0200] Improve the realism and smoothness of the action through optimization algorithms. For example, interpolation and transition techniques are used to make the character's movements look more natural and coherent.

[0201] In summary:

[0202] (1) The present invention constructs a label action library, which contains rich and diverse actions, covering various actions in human daily life, such as walking, running, greeting, raising hands, etc. These actions can cover different situations and interaction needs, from simple basic actions to complex emotional expressions.

[0203] (2) The present invention establishes an effective action library management technology to support the organization, classification and retrieval of actions. In addition, the action library is regularly updated and expanded to adapt to new application requirements and technological advances, and to maintain the timeliness and comprehensiveness of the action library.

[0204] (3) The present invention designs dynamic transition optimization technology to ensure smooth transition between different actions and avoid abrupt or unnatural action switching.

[0205] (4) The present invention applies supervised learning to action generation, which can effectively learn the significant relationship between input data and predefined action labels. This method can not only learn the basic characteristics of the action, such as type, speed and direction, but also identify and learn the variation and complexity of the action, making the model more accurate and reliable in practical applications.

[0206] (5) The present invention has adaptive learning capabilities and can continuously improve and optimize the action generation and selection algorithms based on actual application scenarios and user feedback. This capability enables continuous adaptation to new usage scenarios and interaction requirements, improving user experience and technical competitiveness.

[0207] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for intelligently controlling a digital human using action tags, characterized in that: The steps include: Step S1: Determine the scope and purpose of the action and classify the actions, and clarify the specific application scenarios and target user groups that the action library will be used for; Step S2: Design a technical label system to describe various action features and provide a basis for technical action recognition and analysis; Step S3: Building a behavior control model of the digital character based on the labeled action description; Step S4: Through cross-validation, the data set is divided into multiple subsets for multiple training and validation. According to the evaluation results, the structure of the model is adjusted, the number of layers is increased or decreased, the number of nodes in each layer is adjusted, or the network connection method is changed to better adapt to the data characteristics and the complexity of the problem; Step S5: Intelligent connection and transition of actions: Seamlessly connect the output single actions together to make them look smooth and coherent visually and dynamically without any abrupt or unnatural feeling, and make the action transition smoother by adjusting the position and interpolation method between key frames. Step S6: Unified drive of digital human behavior: It can select appropriate actions according to the current scene, task or user input; it can also dynamically adjust the behavior of the digital character according to the user's input or command, including responding to the user's control instructions, interactive selection or real-time user feedback to adjust the character's action performance.

2. The method for intelligently controlling digital human using action tags according to claim 1, characterized in that: The action library in step S1 includes the following categories:

1. Basic movements: Movement: walking, running, jumping, crawling, sliding, turning; Posture movements: sitting, standing, lying, bending, leaning over, and turning sideways; 2. Compound Actions: Operation actions: open door, close door, open window, close window, take, place, carry; Interactions: handshake, hug, wave, fist, touch, hug; 3. Expressions and emotional movements: Facial expressions: smiling, angry, surprised, sad, confused; Posture expression: confident, anxious, relaxed, nervous, focused; 4. Social and daily life activities: Social actions: greeting, saying goodbye, talking, winking; Daily life activities: eating, drinking, washing, working, and resting; 5. Professional field actions: Medical field: examination, treatment, and surgical operation; Industrial field: assembly, maintenance and operation of machines; Artistic performances: dance movements, drama performances, music performances; 6. Environmental Interaction Actions: Indoor space: entering a room, leaving a room, approaching an object, moving away from an object; Outdoor spaces: through traffic, through people, through natural obstacles; 7. Action attributes: Speed ​​and intensity: fast, slow, gentle, strong; Precision and accuracy: precise, imprecise, high or low accuracy.

3. The method for intelligently controlling digital human using action tags according to claim 1, characterized in that: The technically designed labeling system includes the following steps: Step S2.1: Define the attributes and classification of the tags. First, consider the type of action, such as static or dynamic action. Second, determine the direction of the action, including forward, backward, left, right, and up and down. Finally, mark the speed of the action, from fast to slow. Step S2.2: determining the structure and format of the label, determining the structure and format of the label to describe each action in detail by the type of action (walking, running, jumping, turning), direction (including forward, backward, left, right, up, down, clockwise, counterclockwise), and speed (fast, medium, slow levels); Step S2.3: Design the naming convention of the label; Step S2.4: Consider action combinations and composite labels.

4. The method for intelligently controlling digital human using action tags according to claim 1, characterized in that: The step S3 specifically includes the following steps: supervised learning learns the association of action label sequences from labeled data and is trained with carefully labeled data sets to learn the significant relationship between input data and predefined action labels; the algorithm builds a model by analyzing a large number of labeled examples, which contain multiple variants and scenarios of different actions; by learning these data, the supervised learning algorithm can infer the patterns and regularities of different action features; during the process of model training, the algorithm will continuously optimize its internal parameters to accurately predict the action label sequence in new data to the greatest extent possible.

5. The method for intelligently controlling digital human using action tags according to claim 4, characterized in that: The establishing of the behavior control model further includes feature extraction, and the feature extraction includes the following steps:

1. Data preprocessing: Data cleaning: remove outliers and process missing data; Data smoothing: Use filters or smoothing techniques to smooth data and remove high-frequency noise; Standardization: ensure that the data is on the same scale to avoid the impact of different feature value ranges on the model; 2. Time series feature extraction: If the data is time series, i.e. action sequence or motion capture data, use the following time series features: Statistical characteristics: such as mean, variance, maximum, minimum, standard deviation, etc. These characteristics can describe the distribution and central tendency of data; Time domain features: including the acceleration, angle, speed, etc. of the action. These features can be calculated directly from the original data; Frequency domain features: Frequency domain features are extracted through Fourier transform and other techniques to analyze the periodicity or frequency distribution of actions; 3. Spatial feature extraction: For tasks involving spatial position or posture, such as human action recognition, the following spatial features are used: Joint angle: Calculate the angle from the joint position data; Joint velocity and acceleration: Calculate the joint velocity and acceleration by the rate of change of joint position; Pose descriptor: a representation of human posture, based on keypoint-based posture descriptors or image features; 4. Feature selection and dimensionality reduction: Feature selection: Select the most relevant or representative features, automatically based on the model. Dimensionality reduction: Use principal component analysis (PCA) to reduce the high-dimensional feature space to a lower dimension, retaining the most important information.

6. The method for intelligently controlling digital human using action tags according to claim 1, characterized in that: In step S4, the effect and performance of the model are measured by the following six evaluation indicators:

1. Real-time responsiveness: measures the model's ability to respond immediately to user input, including processing speed and real-time performance; Semantic understanding accuracy: This evaluates the model’s ability to understand user intent, including identifying the topic, sentiment, and specific requirements of the question.

3. Information accuracy: Check whether the information provided by the model is accurate, complete and useful, including the accuracy of facts and the relevance of content; 4. Action generation quality: evaluate the fluency, grammatical correctness and clarity of the actions generated by the model; 5. Behavioral consistency: Check whether the model can maintain consistency in behavior and context over a long period of time; 6. User satisfaction: Measure the user's overall satisfaction with the model interaction experience through user feedback or questionnaires.

7. The method for intelligently controlling digital human using action tags according to claim 1, characterized in that: The specific steps of step S5 are as follows: Step S5.1: Analyze the action sequence: First, the action sequence to be connected needs to be analyzed; Step S5.2: Adjust keyframes: Determine the key frames for each action, including the start frame and the end frame; Step S5.3: Interpolation and curve adjustment: Use appropriate interpolation methods (such as linear interpolation, Bezier curves, etc.) to ensure smooth transitions between keyframes; Step S5.4: Add transition frames: Add transition frames between different actions, which include transition poses and transition actions; Step S5.5: Time interval adjustment: Adjust the time interval between keyframes as needed. Step S5.6: Overall rhythm and logical consistency: Make sure the duration of each action is consistent with the rhythm and logic of the overall animation or game; Step S5.7: Flexible response to scenario requirements: Depending on the needs of the specific scene, you may need to adjust the duration of the action to make the transition between different scenes more natural and seamless.

8. The method for intelligently controlling digital human using action tags according to claim 1, characterized in that: In step S6, selecting an appropriate action according to the current scene, task or user input includes the following specific steps: Step S6.1: Scene recognition and state judgment: Able to accurately identify the current scene and role status of the digital character; Step S6.2: Action selection and sequence creation: According to the recognized character state, an appropriate basic action is selected from a predefined action library; Step S6.3: Environmental Adaptability: When environmental conditions change, the digital character's movements can be dynamically adjusted; Step S6.4: User input and real-time feedback response: Ability to respond to user control commands or interactive selections to adjust the digital character's behavior; this includes real-time user feedback, requiring the ability to instantly update the character's movements to meet user needs and operations. Step S6.5: Action optimization and performance details: Improve the realism and smoothness of actions through optimization algorithms.

Images