Artificial intelligence automatic arrangement method and system suitable for gymnasium dance

Abstract

The invention discloses an artificial intelligence automatic arrangement method and system suitable for gymnasium, and relates to the field of digital image processing, and the method comprises the steps: inputting user information and a music file needing to be used, the user information comprising a fitness target, a physical fitness level, exercise preferences and body parameters of a user; analyzing an input music file, and determining matched actions and starting and ending moments of each action according to rhythm, emotion and tone of music; on the basis of user information, all actions are transitionally coherent through a diffusion model, and the actions are optimized through a physical guiding mechanism and inverse kinematics; and generating an action sequence. According to the method, a more innovative, smoother and diversified action sequence can be generated through the diffusion model, and the problem of repeatability in traditional gymnasium dance arrangement is avoided. Each arrangement can be individually adjusted according to user requirements, and richer exercise experience is provided. And through personalized action arrangement, the requirements of the user can be more accurately met.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of digital image processing, and in particular to an artificial intelligence automatic arrangement method and system suitable for fitness dance. BACKGROUND

[0002] Most current automatic arrangement systems for fitness dance rely on fixed motion libraries and preset rules, and the motion sequences are prone to repetition and lack of creativity. Even if generated automatically through algorithms, some "rigid" motion sequences may still be generated. The motion generation system usually requires human intervention in the connection of motions, especially during multi-motion transitions. The generated results are prone to appear stiff or unnatural, and lack smooth transitions.

[0003] On the one hand, many existing systems cannot generate personalized motion sequences according to the user's personal needs, physical fitness level, health status, and exercise goals (such as fat loss, body shaping, and heart-lung function enhancement). This results in motion arrangement that cannot accurately meet the requirements of different users. Often lacking the ability to adapt to users, especially when the user's physical fitness level and exercise goals change, the system's response is relatively slow or not adaptive.

[0004] On the other hand, the automatic arrangement system for fitness dance often fails to fully consider the biomechanical principles of the human body, generating motions that may exceed the safe range, increase joint burden, or cause unnecessary sports injuries. Lack of reasonable assessment of motion intensity, amplitude, and posture, it is easy to produce motions that are not friendly to the body, especially for beginners or elderly users, lack of appropriate protective measures.

[0005] This is because most existing systems usually rely on single modal input (such as using only music, using only video, etc.), lack the ability to consider multiple modal inputs (such as music rhythm, user description, video, etc.) to generate motion sequences, resulting in limited diversity and adaptability of the generated results. Lack of effective response and adjustment to real-time motion feedback of users. The system usually has no way to provide dynamic adjustment during user execution of motions, resulting in a large gap between the motion and the user's execution.

[0006] Therefore, there is an urgent need to provide a scheme for an artificial intelligence automatic arrangement method and system suitable for fitness dance. SUMMARY

[0007] To solve the above problems, the technical scheme of the present application provides an artificial intelligence automatic arrangement method and system suitable for fitness dance, which can formulate motions according to the user's situation, solving the problem of motion adjustment.

[0008] According to the first aspect of the technical scheme of the present application, an artificial intelligence automatic arrangement method suitable for fitness dance is provided, comprising: S1, input user information and music files to be used, wherein the user information includes the user's fitness goal, physical level, exercise preference and body parameters; S2, analyze the input music files, determine the matching actions according to the rhythm, emotion and tone of the music, and the starting and ending time of each action; S3, based on the user information, make the transition of each action coherent through the diffusion model and optimize the action through the physical guidance mechanism and inverse kinematics; S4, generate the action sequence.

[0009] In the above scheme, step S2 includes: S21, extract the number of beats of the music and the change of the music rhythm, and analyze the emotion and tone of the music; S22, select the corresponding action in the action library according to the rhythm, determine the starting and ending time of each action, and match the emotion change with the action and the tone with the action.

[0010] In the above scheme, step S22 includes: marking the matching points of the action and the music, and corresponding the action and the music beat.

[0011] In the above scheme, step S3 includes: S31, through the diffusion model, multiple iterations of forward diffusion and reverse diffusion are performed to make the transition of the action coherent; S32, through the physical guidance mechanism and inverse kinematics, the action is optimized to conform to the human motion law.

[0012] In the above scheme, step S31 includes: S311, through the diffusion model, gradually add noise to the action data through forward diffusion and gradually remove noise to restore the structure of the action data through reverse diffusion, to generate a sequence that conforms to the target action; S312, according to the user information, adjust the complexity of the action through the diffusion model, including adjusting the amplitude, speed and accuracy of the action; S333, through the diffusion model, multiple iterations are performed to smoothly transition between actions.

[0013] In the above scheme, step S4 includes: displaying the generated action sequence to the user in a visual manner, and the action sequence is displayed including the duration, rhythm and transition point of each action.

[0014] In the above scheme, it also includes that the user can save the generated action sequence and feedback on the generated action sequence.

[0015] In the above scheme, it also includes: real-time monitoring the difference between the user's action and the recommended action and dynamically adjusting the action sequence according to the user's real-time performance.

[0016] According to the second aspect of the technical scheme of the present application, an artificial intelligence automatic arrangement system suitable for fitness dance is provided, which is used to implement the method of any one of the above schemes, and the system comprises: An input module is configured to input user information and music files to be used, wherein the user information includes fitness goals, physical fitness levels, exercise preferences and body parameters of the user; An action arrangement module is configured to analyze the input music files, determine matching actions according to the rhythm, emotion and tone of the music, and determine the start and end time of each action; An action adjustment module is configured to make each action transition coherent based on the user information through a diffusion model and optimize the action through a physical guidance mechanism and inverse kinematics; An action generation module is configured to generate an action sequence.

[0017] According to the third aspect of the technical scheme of the present application, an electronic device is provided, which comprises: A memory storing executable instructions; A processor running the executable instructions in the memory to implement the method of any one of the above schemes.

[0018] The present application has the following advantages: 1. The diffusion model can generate more innovative, smooth and diversified action sequences, avoiding the repetition problem in traditional fitness dance arrangement. Each arrangement can be personalized according to user needs, goals and emotions, providing a richer exercise experience.

[0019] 2. Through personalized and customized action arrangement, the user's needs can be met more accurately. Whether it is simple actions for beginners or challenging actions for advanced users, the system can automatically adapt and provide the best action sequence, thereby enhancing user engagement and satisfaction.

[0020] 3. Through the automatic arrangement and generation system, the cost and time of manual arrangement are greatly reduced. Especially in team courses or online fitness platforms, coaches no longer need to arrange personalized actions for each user, greatly improving work efficiency.

[0021] 4. The physical guidance mechanism and inverse kinematics are used to adjust the action to ensure the safety of the action and reduce the risk of sports injury. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only represent some of the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from the structures shown in the drawings without creative labor.

[0023] Figure 1 A flow chart of the artificial intelligence automatic arrangement method for fitness dance disclosed by the present application.

[0024] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0025] The exemplary embodiments will be described in detail herein with reference to the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present disclosure. Instead, they are merely examples of apparatuses and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0026] The terms "first", "second", and the like in the description and claims of the present disclosure are used for distinguishing between similar objects and do not necessarily have a particular order or sequence. It should be understood that data used with such terms can be interchangeable under appropriate circumstances such that the embodiments of the present disclosure described herein can operate in other sequences than those described or otherwise illustrated herein.

[0027] In addition, the terms "comprise", "comprising", "include", "including", and their conjugates, denote an open-ended inclusion, such that the processes, methods, articles, or apparatuses that include a list of steps or units not necessarily limited to those clearly listed, but can include other steps or units not clearly listed or inherent to such processes, methods, articles, or apparatuses.

[0028] Multiple, including two or more.

[0029] And / or, it should be understood that the term "and / or" used in the present disclosure is only a description of the association relationship of the associated objects, which means that there can be three relationships. For example, A and / or B can represent three cases: A exists alone, A and B exist together, and B exists alone.

[0030] As Figure 1 shown, one embodiment of the technical solution of the present application provides an artificial intelligence automatic arrangement method suitable for fitness dance, comprising: S1, input user information and music files to be used, wherein the user information includes: user's fitness goals, physical fitness level, exercise preferences and body parameters; S2, analyze the input music files, determine the matching actions according to the rhythm, emotion and tone of the music, and the starting and ending time of each action; S3, based on user information, transition each action through diffusion model and optimize action through physical guidance mechanism and inverse kinematics; S4, generate action sequence.

[0031] In step S1, the fitness goals include fat loss, body shaping, and heart and lung function enhancement. These goals will affect the intensity, rhythm and complexity of the actions. For example, fat loss goals may prefer aerobic and high-intensity actions, while body shaping goals may add more strength training actions.

[0032] The physical fitness level determines the complexity and intensity of the actions to be programmed, and the physical fitness level can be quantified by the following formula: 100 (1) The determination of the physical fitness level according to the calculation result includes the following: Beginners have lower physical fitness level, and are recommended to do simple actions with low intensity.

[0033] Intermediate users have higher endurance and are suitable for medium-intensity actions.

[0034] Advanced users need high-difficulty and challenging actions.

[0035] Users can choose their favorite exercise types (such as aerobic, yoga, dance) and exercise intensity (light, medium, high), which helps further customize action sequences and increase user engagement.

[0036] Similarly, the user's height, weight, age and health status, etc. Body parameters are important references for customizing fitness programs for users, especially in terms of action intensity, posture adjustment, etc. Height and weight can affect the amplitude and intensity of actions.

[0037] Of course, in addition to inputting basic information, users can also input other additional information to ensure the accuracy and personalization of action programming, including: Through text description of requirements, such as fitness dance suitable for beginners or actions emphasizing core training, through natural language processing (NLP) technology such as BERT model, semantic analysis of text, and extraction of key action requirements and goals from the text.

[0038] Users can upload reference pictures or videos, which are analyzed by image recognition techniques such as Convolutional Neural Networks (CNN) to extract features like pose, angle, and speed of the fitness movements. These features provide specific movement references for the system, helping it generate similar or consistent movement sequences. By extracting skeletal key points (such as joint positions and angles) from images or videos, the system can better understand the type of movement the user is requesting. Skeletal tracking techniques like OpenPose can extract the positions of various body parts and generate movements that match them.

[0039] Step S2 includes: S21, extract the number of beats and the change of music rhythm, analyze the emotion and tone of the music through a deep learning model; Wherein the deep learning model includes a convolutional neural network CNN, which analyzes the music file and extracts its rhythm, emotion and tone features.

[0040] Extracting rhythm features includes extracting the number of beats per minute (BPM) and the change of music rhythm.

[0041] BPM is an indicator of music speed, which determines the execution speed of each movement. BPM is calculated by analyzing the time domain features of the audio signal, such as the interval time of notes. The calculation formula is: (2) Where T beat is the duration of each beat (seconds). BPM is the basis for determining movement rhythm, higher BPM requires faster movements, and lower BPM is suitable for slow and smooth movements.

[0042] In this embodiment, periodic rhythm patterns are extracted through signal decomposition techniques such as Fast Fourier Transform (FFT) or Short-Time Fourier Transform (STFT). Using periodicity detection algorithms, the system can identify the rhythm pulse in the music, providing rhythm reference for subsequent movement generation.

[0043] Pulse Period = STFT(x(t)) (3) Where x(t) is the audio signal, and STFT is the short-time Fourier transform used to extract the frequency spectrum information of the signal.

[0044] Emotion analysis of audio signals is performed through a convolutional neural network to extract emotional features in the audio. Through an emotion classification model, it can identify whether the music is biased towards happy, sad, inspiring, calm, etc. These emotional types determine the style of movement, for example, music with an inspiring emotion may correspond to more intense and challenging movements.

[0045] Emotion type = CNN(x(t)) (4) where x(t) is the audio signal, and the CNN model is trained to identify the emotion features in the audio.

[0046] Further, the embodiment can quantify the emotion features and express the emotion intensity as a numerical value, so that the system can generate corresponding style actions according to different emotion intensities. For example, the excitement emotion may correspond to intense and dynamic actions, while the calm emotion may require soft and smooth actions.

[0047] (5) where x i is the emotion feature of the audio (such as pitch, loudness), w i is the weight, n is the number of features, is the emotion intensity.

[0048] Through pitch tracking of the audio signal, the system can extract the trajectory of the pitch over time. Commonly used pitch tracking algorithms include the autocorrelation function (ACF) and YIN algorithm. The pitch information can affect the smoothness and variation of the action.

[0049] P(t) = Pitch(x(t)) (6) where P(t) is the pitch of the audio signal at time t, and x(t) is the audio signal.

[0050] S22, according to the rhythm, select the corresponding action in the action library, determine the starting and ending time of each action, and match the emotion change and the action, the pitch and the action.

[0051] In this embodiment, according to the BPM (music rhythm) obtained by the above analysis, the starting and ending time of each action is determined. Usually, the system will insert actions at the rhythm points of the music (for example, at the beginning of each measure or at the time of the repeat), to ensure that the action is consistent with the rhythm of the music.

[0052] Assuming that the BPM of the music is XBPM, the starting time of the action is determined according to the length of each rhythm: (7) where current beat is the beat number of the current music, and BPM is the number of beats per minute of the music.

[0053] According to the change of emotion intensity, the system adjusts the style of the action at the time of strong emotion change. For example, when the emotion changes from calm to excitement, the complexity and intensity of the action will increase. The system will generate new actions at the threshold of emotion change.

[0054] (8) When the change in emotional intensity is greater than the threshold, the system adjusts the intensity of the action according to the change.

[0055] Further, the embodiment matches the pitch of the tone with the fluctuation of the action. For example, when the tone rises, the amplitude of the action increases, and when the tone decreases, the intensity of the action decreases. The system adjusts the intensity and type of action through the frequency and amplitude of the pitch change.

[0056] Assuming the pitch change is Pchange, the system adjusts the amplitude of the action according to the change in pitch: A action = P change ×k (9) Where A action is the amplitude of the action, P change is the amplitude of the pitch change, and k is the adjustment factor.

[0057] Step S3 includes: S31, through the diffusion model, multiple iterations of forward diffusion and backward diffusion, make the action transition coherent; S311, gradually add noise to the action data through forward diffusion and step-by-step denoising to recover the structure of the action data through backward diffusion, to generate a sequence that meets the target action; S312, adjust the complexity of the action according to the user information through the diffusion model, including adjusting the amplitude, speed and accuracy of the action; S333, through the diffusion model multiple iterations, smooth transition between each action.

[0058] Diffusion Models is a generative model that generates data through a process of gradually introducing noise and then gradually denoising. It converts noise into structured data (such as action sequences) through multiple iterations, showing great advantages in generating coherent action transitions and high-difficulty action connections. The core idea is to restore the target data (action sequence) by gradually "denoising" from the initial high-noise state. Each denoising process is carried out through a neural network model, and finally the clear sample of the target is recovered from the noise.

[0059] In the forward diffusion process, noise is gradually added to the original data (such as action data), causing the data to gradually lose structure. Assuming the initial data is x0, noise is added to the data through multiple steps: (10) Where x t is the data after t steps of noise addition, is the noise addition ratio of each step, is the random noise.

[0060] The reverse diffusion process restores the structure of the data by step-by-step denoising. In action generation, the reverse process is learned by a trained neural network to map from noise to data: (11) where, is the output of a neural network model, which predicts the noise at time step t based on the current noise state and removes the noise.

[0061] In this process, the network gradually generates output that matches the target action sequence by continuously adjusting parameters. Each round of denoising brings a more accurate action sequence.

[0062] Further, the diffusion model can also combine the input user information to adjust and generate action sequences suitable for the user.

[0063] On the one hand, according to the user's physical level, the diffusion model will adjust the complexity of the generated action sequence. For example, beginners will generate relatively simple actions, while advanced users will generate complex and high-intensity actions. This matching can be quantified by the following formula: (12) where, is the action complexity, is the user's physical level (beginner, intermediate user or advanced user), and the system adjusts the generated action amplitude, speed and accuracy to match the user's physical ability.

[0064] On the other hand, according to the user's target requirements (such as fat loss, shaping, etc.), generate actions that match the target. For example, fat loss goals may generate more aerobic and high-intensity actions, while shaping goals focus on strength training, action stability and control.

[0065] G action =Goal(x target ) (13) where G action is the action feature related to the target, and x target is the user's fitness goal (such as fat loss, shaping, etc.). Through multiple iterations in the diffusion process, the system can gradually explore the best transition point of each action and reduce the abruptness between actions. For example, when high-difficulty action transitions are needed, the system will generate smooth transition actions based on factors such as music rhythm, emotional changes, etc.

[0066] The smoothness of the transition can be measured by a similarity metric between actions. For example, the Euclidean distance is used to calculate the similarity between two actions, and this distance is minimized during the generation process: (14) where, is the Euclidean distance between the i-th and i+1-th actions, denotes the position of the j-th key point in the i-th action.

[0067] The present application can generate complex and coherent action sequences using an enhanced diffusion model, especially when high precision action connection is required (such as the connection of multiple difficult actions). Through the diffusion process, the system can automatically explore the optimal transition points between actions, reducing stiff or abrupt action switching. At the same time, the diffusion model can generate personalized actions according to the user's body parameters, music rhythm, and target requirements, with high randomness and innovation, avoiding the repetition and singularity of action choreography. The enhanced diffusion model includes multi-modal generation fusion, using a cross-modal architecture (such as the CLIP model) to map features of different modalities (text, music, images) into the same latent space, and combining the diffusion process to generate high-quality action sequences. A physical guidance mechanism is added to generate actions that are more consistent with human motion laws. With the help of an inverse kinematics (Inverse Kinematics) model, the generated actions are aligned with the real human motion model. In the diffusion process, constraint-based sampling is added to guide the action generation process. A meta-learning (MAML) framework based on gradient is embedded in the diffusion model, allowing the model to quickly adapt to the data distribution of new users after training.

[0068] S32, optimize the action through the physical guidance mechanism and inverse kinematics to conform to the human motion law.

[0069] Ensure that the generated actions conform to the human motion law. For example, the system will automatically adjust the action amplitude and direction to avoid producing excessive stretching or actions that are not suitable for some users.

[0070] One of the core advantages of the diffusion model is its random generation capability, i.e., each generated action sequence is unique, avoiding the repetition and singularity of actions. By introducing noise and randomness, the system can explore new action patterns in each action generation, providing more creative choreography.

[0071] Specifically, when the user provides multiple modal inputs such as text, pictures, or videos, the system can use multi-modal learning technology to fuse the information of these modalities into the action generation process, thereby enhancing the diversity and accuracy of action choreography.

[0072] The user-provided textual description, such as "fitness dance suitable for beginners" or "movements emphasizing core training", is converted into usable movement features through natural language processing (NLP) techniques, guiding the style and complexity of the movements.

[0073] T action = NLP(x text input ) (15) The user-provided reference images or videos are processed by a convolutional neural network (CNN), extracting features such as skeleton posture, angles, and speed of the movements, and incorporating these features into the diffusion model to generate movements that better match the reference images or videos.

[0074] I action = CNN(x mage / video input ) (16) Through multi-modal fusion, the system can more accurately understand user needs and generate movements that better meet user expectations.

[0075] Step S4 includes: presenting the generated movement sequence to the user in a visual manner, and the movement sequence is presented including the duration, rhythm, and transition points of each movement.

[0076] Users can view the complete fitness dance movement process as needed, or choose a single movement for learning and adjustment. According to the actual experience feedback of the system's movement generation effect, the system will further optimize the movement arrangement according to user feedback. At the same time, the generated movement sequence and training plan will be saved to the user's personal profile for the user to review or reuse at any time.

[0077] In the output stage, the system will present the processed and generated movement sequence and provide a feedback mechanism, allowing users to make adjustments according to actual needs, and finally save and apply them to their training process. The goal of this stage is to ensure that users can clearly and effectively understand and apply the arranged movement sequence, while providing feedback for system optimization.

[0078] Specifically, the presentation of the movement sequence includes the following aspects: Movement duration: the execution time of each movement. The system calculates the duration of each movement based on the rhythm of the movement and the user's physical fitness requirements. Assuming the duration of the movement is T action , where T action is calculated by the following formula: (17) Where BPM is the number of beats per minute (music rhythm), duration per beat is the duration of each beat, and number of beats in action is the number of beats included in the movement.

[0079] Rhythm of Action: The rhythm of each action is automatically adjusted according to the rhythm of the music and the user's preferred target, ensuring that the execution rhythm of the action is consistent with the music and user needs. The system determines the rhythm R of each action based on the rhythm (BPM) of the music and the dynamic requirements of the action action , i.e. the rate of action, defined as the number of times the action is performed per unit of time: (18) Transition Points: The system identifies transition points between actions based on the naturalness and smoothness of action transitions. These transition points are critical moments in the action sequence that enable smooth transitions. The transition points are calculated based on the output of the diffusion model, which optimizes the smoothness of the transition by calculating the transition error between actions. The formula for calculating the transition point can be expressed as: (19) where, is the transition point of action transition, and are the start and end times of two consecutive actions.

[0080] Through the graphical user interface, the system visualizes the actions and rhythms, and the user can choose to view the details of each action (such as the angle of action execution, posture, etc.). For example, the user can view an animation showing the execution process of each action, or according to their own needs, view a key posture breakdown of the action.

[0081] Example demonstrations include the following: The user views the complete dance action sequence, including the duration, rhythm, and transition points of each action. By clicking on a specific action, the system displays the detailed execution steps of that action, including joint angles, body movement trajectories, and other details. The user can choose to adjust the action duration, rhythm, and other parameters, and the system automatically updates the action display according to the new input.

[0082] In the output stage, the system also provides feedback and saving functions to optimize according to the user's actual experience and ensure that the action sequence can be used and consulted for a long time.

[0083] The system allows users to provide feedback on the generated action sequence. User feedback typically includes the following aspects: Action Difficulty: User feedback on whether the action is too difficult or too easy.

[0084] Action Smoothness: User feedback on whether the action is smooth and the transition is natural.

[0085] Personalized Suggestions: Users can make personalized suggestions on action intensity, rhythm, etc.

[0086] The system integrates the user's feedback information through the following formula and adjusts the motion choreography according to the strength of the feedback: (20) Where △A represents the adjustment amount of the motion choreography, is the user's feedback on the i-th motion, is the weight of each feedback, and n is the number of feedbacks given by the user. The system calculates the total feedback by weighted average and adjusts it when generating the next round of motion.

[0087] After the system generates the final motion sequence and training plan, the user can choose to save it in the personal profile for easy access or reuse. The saved data can include the following: Motion sequence: detailed information of each motion, including duration, rhythm, transition points, motion intensity, etc.

[0088] Training plan: personalized training plan generated based on user goals and feedback, including daily or weekly training tasks.

[0089] User personal profile: contains user's historical records, training achievements and personal health data.

[0090] When saving the motion sequence and training plan, the system will generate a unique identifier UID for each user and associate the motion sequence and training plan with the user. Assuming that the user's personal profile data is D, the data T of saving the motion sequence and training plan is represented by the following formula: T u = {action sequence, training plan, feedback history} (21) The user can access T u through the interface to view their historical training data, adjust or reuse previous motion choreography and training plans.

[0091] The user can access T u through the interface to view their historical training data, adjust or reuse previous motion choreography and training plans.

[0092] After each user completes the training, the system will update their training data and use these data to further optimize the motion generation model. For example, the user's feedback and performance data in multiple training sessions can be used to fine-tune the diffusion model, motion generation algorithm or personalized adaptation strategy to improve the accuracy and response speed of the system.

[0093] The embodiment also includes real-time monitoring of the user's actions and the difference between the recommended actions and dynamically adjusting the action sequence according to the user's real-time performance.

[0094] The embodiment also includes using bone tracking technology (such as OpenPose or MediaPipe) to monitor the difference between the user's actions and the recommended actions in real time. For example, by capturing the user's joint positions in real time, analyzing the action completion, and comparing it with the ideal action. By constantly monitoring the user's movement progress and performance, timely adjustment suggestions or new action sequences are provided to ensure that the user always stays within the appropriate training interval. Improve user experience and ensure the scientificity and effectiveness of training.

[0095] The system can also dynamically adjust the action sequence according to the user's real-time performance, including the following: Adjust the action intensity and rhythm. If the user deviates in a certain action, the system can reduce the difficulty of the action or adjust the rhythm to make it easier to execute.

[0096] Personalized recommendations. Based on feedback, the system will provide personalized recommendations, such as "slow down a bit" or "increase the action amplitude".

[0097] Specifically, first, the user's joint positions are obtained through bone tracking technology. At the same time, the system also determines the expected joint positions through the recommended ideal action. In the process of real-time feedback, the system calculates the difference between the user's current action and the recommended action. This difference can be measured by calculating the Euclidean distance of each joint position: (22) Where d j is the difference of the jth joint, representing the deviation between the user's action and the recommended action. The system calculates the overall action error based on the difference of each joint: (23) Where D is the average error of the entire action. If D exceeds the preset threshold, the system will provide feedback to prompt the user to adjust the posture.

[0098] Based on the calculated error D, the system can make corresponding feedback. If the error is large, the system can prompt the user to "slow down a bit" or "increase the action amplitude" to help correct the action. If the error is small, the system will confirm that the user's action is correct and encourage them to continue.

[0099] In the dynamic optimization phase, the system automatically adjusts the user's action sequence based on the results of real-time feedback. This phase includes adjustments to action intensity, rhythm, and action itself to help the user better execute the action.

[0100] According to the difference D of feedback, the system will adjust the intensity and rhythm of the action. If the user's action performance is poor (such as a large error), the system can make the action easier to perform by reducing the difficulty of the action or adjusting the rhythm. The system can also adjust the rhythm of the action to adapt to the user's performance. For example, when the user performs poorly, the rhythm can be slowed down to give the user more time to complete each action.

[0101] According to real-time feedback, the system can provide personalized suggestions for the user. For example, if the user's action amplitude is too small, the system will prompt "increase the action amplitude slightly". If the user's action is too fast, the system can prompt "slow down to maintain the correct posture". These personalized suggestions are generated by calculating the error D of the action and combining the user's physical fitness level, health status and target demand.

[0102] The system will monitor the user's training intensity and adjust the training plan according to the user's performance. If the user's training progress exceeds the expected (such as completing the training of a certain stage too early), the system can continue the training by generating new action sequences; if the user's progress is slower, the system can appropriately reduce the difficulty or adjust the training content to ensure that the user is always within a reasonable training intensity range.

[0103] On the other hand, the present application uses gradient-based meta-learning (such as MAML framework Model-Agnostic Meta-Learning), which can quickly adapt to the needs of new users. After receiving the data of new users, the model quickly adjusts the parameters through a small number of training samples, thereby generating action sequences that meet the needs of new users.

[0104] The training process of MAML includes two main stages: inner loop (Inner Loop) and outer loop (Outer Loop). In the inner loop, the model uses the data of new users for a small amount of training (i.e. one or a few gradient updates). In the outer loop, the model is trained through the loss function of multiple tasks to optimize the initial parameters 0, so that the model can adapt to different needs through a small number of updates when encountering new tasks. Through the MAML framework, the system can quickly adapt to a small amount of new user data (such as physical fitness level, health status, target demand, etc.) and optimize the parameters of the model, thereby generating action sequences that meet the needs of new users. If the new user inputs some data about the physical fitness level and the target demand, the system quickly adapts to the user's characteristics through the MAML framework, and adjusts the generation of action sequences through a small number of gradient updates.

[0105] In addition, personalized adjustments can also be made according to the user's level, including: According to the user's physical fitness level, the system will automatically adjust the intensity and difficulty of the movements. Beginners: the system generates simple movement sequences with low amplitude and speed to avoid excessive consumption or injury. Intermediate users: the system generates more complex movement sequences with some added challenge, but still within the user's physical endurance range. Advanced users: complex and high-intensity movement sequences with higher challenge, including some high-tech and coordinated movements.

[0106] Further, users can provide feedback after completing the training, including the difficulty, fluency, and personal feelings of the movements. Based on a large amount of user feedback data, the system will optimize itself, constantly adjusting and improving the movement generation model, and improving the quality and personalization of the generated movement sequences.

[0107] In the final stage, the system evaluates and optimizes the entire aerobics choreography process through user feedback, and uses this feedback data to continuously optimize itself, ultimately achieving long-term adaptation and personalized customization to user needs. This stage is the key to closed-loop optimization, ensuring that each training data feedback can provide more accurate basis for the next round of choreography and generation.

[0108] According to the second aspect of the technical scheme of the present application, an artificial intelligence automatic choreography system suitable for aerobics is provided, which is used to realize the above method. The system includes: An input module for inputting user information and music files to be used, wherein the user information includes the user's fitness goals, physical fitness level, exercise preferences and body parameters; A movement arrangement module for analyzing the input music file, determining the matching movements according to the rhythm, emotion and tone of the music, and the starting and ending time of each movement; A movement adjustment module for adjusting each movement based on user information through a diffusion model; A movement generation module for generating movement sequences.

[0109] According to the third aspect of the technical scheme of the present application, an electronic device is provided, which includes: A memory storing executable instructions; A processor running the executable instructions in the memory to implement the above method.

[0110] It should be noted that, in this document, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a" does not, without more constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0111] The above-mentioned embodiment numbers of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0112] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned implementation methods can be realized by means of software and the necessary general hardware platform, and of course, they can also be realized by hardware, but in many cases, the former is a better implementation method. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as a ROM / RAM, a magnetic disk, or an optical disk) and includes a plurality of instructions for causing a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the methods described in the various embodiments of the present application.

[0113] The embodiments of the present application are described above in combination with the drawings, but the present application is not limited to the above-mentioned specific embodiments, and the above-mentioned specific embodiments are only illustrative and not restrictive. Those skilled in the art can make many forms under the inspiration of the present application without departing from the purpose of the present application and the scope protected by the claims, and these all belong to the protection of the present application.

Claims

1. An artificial intelligence-based automatic choreography method suitable for aerobics dance, characterized in that, include: S1. Input user information and the music file to be used. The user information includes: the user's fitness goals, fitness level, exercise preferences and body parameters. S2. Analyze the input music file, and determine the matching actions, as well as the start and end times of each action, based on the rhythm, emotion, and pitch of the music. S3. Based on user information, the diffusion model makes the transitions between actions coherent and optimizes the actions through physical guidance mechanisms and inverse kinematics. S4. Generate action sequence.

2. The artificial intelligence automatic choreography method for fitness dance according to claim 1, characterized in that, Step S2 includes: S21. Extract the number of beats and changes in rhythm of the music, and perform emotional and tonal analysis on the music; S22. Select the corresponding action from the action library according to the rhythm, determine the start and end time of each action, and match the emotional changes with the actions, and coordinate the tone with the actions.

3. The artificial intelligence automatic choreography method for fitness dance according to claim 2, characterized in that, Step S22 includes: marking the matching points between actions and music, and matching actions with music beats.

4. The artificial intelligence automatic choreography method for fitness dance according to claim 1, characterized in that, Step S3 includes: S31. Through multiple iterations of forward and reverse diffusion using a diffusion model, the transition of actions becomes smoother. S32. Optimize movements through physical guidance mechanisms and inverse kinematics to conform to the laws of human movement.

5. The artificial intelligence automatic choreography method for fitness dance according to claim 4, characterized in that, Step S31 includes: S311. The motion data is gradually denoised by forward diffusion and backward diffusion to restore the structure of the motion data in order to generate a sequence that conforms to the target motion. S312. Adjust the complexity of the action based on user information using a diffusion model, including adjustments to the amplitude, speed, and precision of the action; S333. Through multiple iterations of the diffusion model, the transitions between various actions are smoothed.

6. The artificial intelligence automatic choreography method for fitness dance according to claim 1, characterized in that, Step S4 includes: displaying the generated action sequence to the user in a visual manner, wherein the action sequence displays the duration, rhythm and transition points of each action.

7. The artificial intelligence automatic choreography method for fitness dance according to claim 1, characterized in that, Also includes: Users can save the generated action sequences and provide feedback on them.

8. The artificial intelligence automatic choreography method for fitness dance according to claim 1, characterized in that, Also includes: It monitors the difference between user actions and recommended actions in real time and dynamically adjusts the action sequence based on the user's real-time performance.

9. An artificial intelligence-based automatic choreography system suitable for aerobics dance, characterized in that, The system is used to implement the artificial intelligence automatic choreography method for fitness dance as described in any one of claims 1-8, and the system comprises: The input module is used to input user information and the music file to be used. The user information includes: the user's fitness goals, physical fitness level, exercise preferences and body parameters. The motion choreography module analyzes the input music file and determines the matching motions, as well as the start and end times of each motion, based on the rhythm, emotion, and pitch of the music. The motion adjustment module is used to make the transitions of various actions coherent based on user information through a diffusion model and to optimize the actions through a physical guidance mechanism and inverse kinematics. The action generation module is used to generate action sequences.

10. An electronic device, characterized in that, The electronic device includes: Memory, which stores executable instructions; A processor that executes the executable instructions in the memory to implement the method of any one of claims 1-8.

Images