Dance treatment posture correction method based on biological feedback

Abstract

The invention relates to a dance treatment posture correction method based on biological feedback, and belongs to the technical field of intelligent rehabilitation and action behavior correction. According to the method, posture data and physiological parameters of a dance participant are synchronously collected through a wearable device, a posture deviation evaluation model is constructed by using a deep neural network, and a physiological state weight adjusting function is constructed by combining heart rate, skin electricity and electromyographic signals; multi-mode feedback control instructions such as directional vibration, voice prompt or augmented reality visual guidance are dynamically generated and act on key joint parts in real time, and high-precision and self-adaptive correction of dancing postures is achieved. The system has an individualized learning function, can update an individualized model based on historical data, and improves the accuracy and robustness of attitude intervention. The method can be widely applied to scenes such as dance training, autism rehabilitation and nerve injury posture reconstruction, and has the advantages of being high in response speed, scientific in feedback mechanism, high in adaptability and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bio-information processing and dance therapy, and particularly relates to a dance therapy posture correction method based on biofeedback, which is suitable for posture behavior correction systems in scenes such as rehabilitation medical treatment, dance teaching and psychological intervention. BACKGROUND

[0002] Dance / movement therapy (DMT) as a comprehensive intervention method combining artistic expression and body movement has been widely applied in the fields of psychological treatment, neurological rehabilitation and cognitive disorder intervention in recent years. Its core idea is to stimulate individual emotional experience and self-cognition through body movement, rhythm perception and posture expression, and then realize the dual improvement of physiology and psychology. Especially in the population of autism spectrum disorder (ASD), Parkinson's disease, stroke rehabilitation and post-traumatic stress disorder (PTSD), dance therapy shows good feasibility and adaptability due to its non-verbal, low threshold and strong interactive characteristics.

[0003] However, the existing dance therapy practice mainly relies on artificial guidance, that is, a dance therapist with professional experience observes and judges the movement behavior, body language and non-verbal feedback of the participants, and then implements personalized oral guidance and demonstration. Although this method has certain flexibility and humanistic care, its shortcomings are also very significant: first, since posture judgment mainly depends on the experience and visual observation of the therapist, there is a lack of quantitative standard and objective evaluation mechanism, which is easy to cause diagnostic errors and inaccurate action guidance; second, the feedback in the treatment process is lagging, especially in the synchronous training of multiple people or complex dance movements, the therapist is difficult to pay attention to the posture changes of all individuals at the same time; in addition, the participants themselves cannot perceive the action error in real time, which is not conducive to the formation of correct body memory and internalization of rehabilitation behavior. Therefore, it is urgent to introduce quantitative technical means into the dance therapy system to realize high-precision, personalized and real-time posture correction and feedback.

[0004] In recent years, the rapid development of gesture recognition and physiological monitoring technology provides technical support for the solution of the above problems. Motion tracking devices based on inertial measurement units (IMU) and skeleton extraction algorithms based on visual recognition have been widely used in sports training and intelligent rehabilitation fields, which can realize high-precision acquisition and three-dimensional modeling of human main joints and limb trajectories. At the same time, the convenient acquisition of physiological signals such as heart rate (HR), galvanic skin response (GSR), and electromyography (EMG) makes it possible to dynamically evaluate the individual's internal state such as tension, emotional fluctuation, and muscle load during exercise. This kind of technology provides a new quantitative window for the "behavior-psychological" coupling in the process of dance therapy.

[0005] However, the current joint analysis of physiological parameters and action posture data and real-time intervention of dance posture is still in its infancy. On the one hand, existing rehabilitation systems are mainly based on physiological monitoring, lacking the ability to recognize and evaluate complex dance action sequences; on the other hand, traditional motion capture systems still have problems such as calibration difficulty and poor real-time performance when facing high-degree-of-freedom and diverse dance actions, making it difficult to meet the dual demands of feedback timeliness and action expression diversity in dance therapy. In addition, although some research has attempted to introduce AI algorithms to model and evaluate dance actions, most systems do not consider the impact of physiological state on posture control strategy. For example, in a specific dance posture action, the individual's performance under psychological stress or muscle fatigue state will be significantly different from that under normal state, and if the control system cannot perceive and adapt to this physiological change, the feedback mechanism will have obvious "over-positivity" or "over-slow" phenomenon, which will affect the correction effect and user experience.

[0006] Furthermore, there is still a lack of system architecture and algorithm system that effectively integrates action deviation modeling, physiological state perception, and individualized feedback strategy. How to realize the coupling modeling of action data and physiological data in the time scale and semantic level, how to dynamically adjust the feedback intensity and mode (such as voice prompt, tactile vibration) according to the current state of the participant, and how to improve the adaptability and generalization ability of the system through individualized learning mechanism, have become the key bottleneck restricting the intelligentization process of dance therapy technology. In summary, the existing dance therapy posture intervention methods have many shortcomings such as strong subjectivity, feedback lag, lack of dynamic adaptation, and lack of an integrated, multi-modal, and intelligent feedback system to adapt to individual differences and dynamically adjust the intervention strategy. Therefore, there is an urgent need for a dance posture correction technology framework based on biological feedback, which integrates motion capture, physiological monitoring, posture deviation modeling, and dynamic feedback mechanism, realizes the technical closed loop of "controllable precision, traceable path, and evaluable effect" for dance therapy, and provides effective support for the standardization, individualization, and popularization of dance therapy. SUMMARY

[0007] The dance therapy posture correction method based on biofeedback aims to solve the problems of strong subjectivity of posture guidance, lagging feedback mechanism, and lack of personalized adaptive strategies in traditional dance therapy. By constructing a closed-loop system integrating multi-modal data acquisition, posture deviation modeling, physiological state quantification, feedback regulation control, and individual adaptive learning, the method realizes accurate identification, evaluation, and real-time intervention of dancer posture state, and improves the effectiveness, scientificity, and intelligent level of dance therapy intervention.

[0008] To achieve the above object, the present application provides the following technical scheme: a dance therapy posture correction method based on biofeedback, characterized in that it comprises the following steps:

[0009] Step S1: dance motion and physiological signal synchronous acquisition, a wearable perception module synchronously acquires posture data and physiological signal data of a dance participant during the treatment process, the posture data includes but is not limited to three-dimensional space coordinates of each skeletal joint, joint angle, limb acceleration, etc., the physiological signals include heart rate (HR), galvanic skin response (GSR), and electromyographic signal (EMG), the sampling frequency can be adjusted according to specific application requirements, and the recommended value is above 50 Hz for posture channels and above 20 Hz for physiological channels. The sensor arrangement refers to the principles of ergonomics, and the key nodes include both shoulders, both knees, the waist, the wrist, and the front of the chest.

[0010] Step S2: standard posture model construction and dynamic alignment, the collected posture data P real (t) is frame-aligned and time-synchronized with the preset standard motion sequence P ref (t), time domain normalization is completed by using a dynamic time warping algorithm or a time convolution network, and a posture difference function is constructed: ΔP(t)=P real (t)-P ref (t), wherein ΔP(t) is a posture error vector at time t.

[0011] Preferably, to enhance the adaptability to different dancers' body types, the standard posture model is constructed by using a parameterized human model, and is adjusted by fusing a human body proportion calibration and a gender and age database.

[0012] Step S3: posture deviation evaluation and grade determination, the posture deviation is input into a pre-trained posture evaluation network PEN (Posture Evaluation Network), which is a time series multi-channel evaluation model based on the Transformer structure, and outputs a posture deviation grade score ∈(t) for each frame, ranging from 0 to 1, where 0 represents complete matching and 1 represents serious deviation.

[0013] Preferably, the network is supervised by training with biomechanics annotation dataset, which can identify typical error types such as spinal deviation, pelvic rotation, knee joint inward buckling, etc.

[0014] Step S4: physiological state feature extraction and weight factor calculation, filter, normalize and extract features from physiological signals such as heart rate HR(t), skin electricity GSR(t) and muscle electricity EMG(t) to obtain physiological state vector:

[0015] B(t)=[N(HR(t)),N(GSR(t)),N(EMG(t))]

[0016] Where N(·) is a normalization function, and the dimension range is unified to [0, 1]. Further define the physiological state weighting function:

[0017] W(t)=α·N(HR(t))+β·N(GSR(t))+γ·N(EMG(t))

[0018] Where α+β+γ=1, which can be adjusted according to the dancer's constitution. W(t) reflects the individual's current tension level, physical load level or emotional stability.

[0019] Step S5: feedback control function generation and multi-mode feedback output, combine posture deviation ΔP(t) and physiological state weight W(t) into feedback control module to generate posture correction instruction:

[0020] C(t)=f(ΔP(t),W(t))

[0021] The feedback control function f(·) is a nonlinear mapping relationship, considering error direction, amplitude, physiological tension state, historical correction record, etc. The final control output is realized through the following multiple ways:

[0022] Wearable vibration module: start when the posture deviation exceeds the set threshold and the physiological state is stable, accurately positioning the vibration feedback part;

[0023] Real-time voice prompt system: provide concise and positive semantic guidance, such as "right shoulder relaxation", "slightly squat", "upper limbs open";

[0024] Augmented reality guidance: superimpose the standard posture skeleton model through AR glasses for the dancer to imitate.

[0025] Step S6: Individual Database Establishment and Dynamic Model Update. The system continuously collects historical data from the participants' training process to build an individualized database. Using deep transfer learning, the system trains an individualized posture evaluation model and feedback strategy module based on a small sample learning strategy. As dancers practice more, the system gradually optimizes their posture evaluation criteria and dynamically updates the reference model, achieving a transition from a "universal model" to a "personalized model."

[0026] Preferably, the Bayesian optimization parameter adjustment method is used to model the individual feedback responsiveness, and the feedback acceptance factor R(t) is set, and the feedback rhythm is adjusted in combination with the actual posture improvement speed.

[0027] The preferred embodiment further illustrates:

[0028] A posture-supporting feedback strategy for Parkinson's disease patients: This group experiences decreased muscle control and significant movement delays. When the system detects persistent posture deviation ΔP(t) and abnormal myoelectric output W(t), it automatically reduces feedback frequency and extends response time to avoid mental strain.

[0029] Collaborative learning mechanism for youth dance training: supports duet dance interaction mode, determines movement coordination through the posture vector difference value of the two wearable systems, and provides synchronous vibration feedback on the core joints of both (such as handshake position and shoulder line parallelism).

[0030] Emotion-guided intervention for children on the autism spectrum: If the system detects a continuous decrease in GSR (indicating decreased attention or emotional withdrawal), the difficulty of the original movement will be reduced and switched to a standard dance posture model with a slower rhythm. At the same time, the voice encouragement module will be activated, outputting sentences such as "You are doing well, keep it up."

[0031] The present invention has the following beneficial effects:

[0032] 1. Quantitative assessment of posture errors: By combining the PEN deep neural network with a standard motion database, the system can identify subtle differences in posture at the frame level, compensating for the subjectivity and large error range of traditional manual observation methods.

[0033] 2. Construct a "movement-physiology" coupled feedback mechanism: Using physiological state as a feedback regulator, dynamically control the correction rhythm and feedback intensity to achieve flexible intervention for dancers who are under high pressure, prone to fatigue, or have emotional fluctuations.

[0034] 3. Multimodal feedback improves training efficiency: Unlike a single voice prompt system, wearable vibration feedback has the advantages of being intuitive, private, and having a short response time. It is particularly suitable for clinical rehabilitation and individualized dance training scenarios.

[0035] 4. Learning ability and adaptability: The method runs with a standard general model in the initial stage of collection, and then gradually adjusts the feedback strategy with the support of individual database, to realize the gradual adaptation and optimization of body condition, reaction time and emotional state. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present application will be described clearly and completely below. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application. The present application provides a dance therapy posture correction method based on biofeedback, mainly including:

[0037] Multi-channel acquisition module: realizing posture data and physiological signal synchronous acquisition;

[0038] Posture evaluation network PEN: for posture error calculation and grade evaluation;

[0039] Physiological state weight calculation unit: for real-time extraction of individual physiological state characteristics;

[0040] Feedback control module: for generating vibration / voice / AR feedback;

[0041] Personalized learning and model updating module: realizing continuous optimization of the system personalized correction strategy.

[0042] 1. Multi-modal data synchronous acquisition, the wearing device includes six nine-axis IMU sensors (for head, shoulders, waist, knee joints), two electromyographic sensing electrodes (for biceps femoris and triceps), one heart rate and GSR composite sensor (in front of the chest). The sampling period T is set to 20 ms, the data recording time window is T = 5 s, and the collection sequence in each period is: s w

[0043] D(t) = {P real (t), HR(t), GSR(t), EMG(t)}, t ∈ [0, T w ] (1)

[0044] The posture signal is constructed as a bone joint angle sequence with IMU inertial data, and a Madgwick filter is used for posture fusion to obtain the joint posture vector Preal(t) at each time.

[0045] 2. Posture deviation modeling, taking the standard dance movement "arabesque" as an example, the standard joint sequence is represented as Pref(t), and the deviation is:

[0046] ​​

[0047] where, represents the angle deviation of the ith joint, n is the number of key bones (such as shoulder, elbow, knee, ankle, waist, a total of 10 degrees of freedom). Calculate the Euclidean norm of the overall pose error:

[0048]

[0049] Further input into the trained PEN model, output the pose deviation level score of each frame, and the level mapping is shown in the following table 1.

[0050] Table 1 Pose deviation level score and correction strategy

[0051]

[0052] 3, Physiological state weight modeling, the HR, GSR and EMG signals collected synchronously are wavelet filtered to remove motion artifacts, and then standardized:

[0053]

[0054] Then the physiological state weight is:

[0055] W(t) = aHR * + βGSR * + γEMG * (5)

[0056] Set the weight parameters a = 0.4, β = 0.3, γ = 0.3. When W(t) > 0.7, it indicates that the dancer is in a state of tension, and the system automatically delays the feedback frequency and weakens the vibration intensity.

[0057] 4, Feedback control instruction generation, define the feedback control function as:

[0058] C(t) = δ1·(t)·V haptic + δ2·(t)·W(t)·S voice (6)

[0059] It inputs the historical error sequence into the TCN network for personalized modeling, and outputs the optimal path of the feedback strategy at T+1 time in the future:

[0060]

[0061] The system will error back-propagate the predicted feedback strategy and the actual correction effect every time the training iteration is performed, optimize the TCN weight, and make the model's adaptability to the individual increase with the number of uses. Through field verification, the response ability and posture correction effect of the test system under different action complexity and psychological state are tested. Two typical dance poses, "Arabesque" and "Swan Arm", are selected for comparative experiments, and the experimental data are shown in Table 2.

[0062] Table 2 Change in posture error before and after correction

[0063]

[0064] The improvement rate is defined as:

[0065]

[0066] The results show that the method of the present application can significantly improve the dance pose accuracy while ensuring the stability of the trainer's emotions, the feedback delay is controlled within 100 ms, and the system can realize quasi-real-time feedback.

Claims

1. A dance therapy posture correction method based on biofeedback, characterized in that: The steps include: Step S1: collecting multi-channel physiological and posture data of the dance participants, wherein the posture data includes information on joint angles and skeletal center of gravity trajectory information, and the physiological data includes heart rate HR, galvanic skin response GSR, and electromyographic signal EMG; Step S2: construct a time sequence alignment between the standard posture sequence Pref(t) and the actual collected posture Preal(t) to obtain the posture deviation vector ΔP(t)=Preal(t)-Pref(t); Step S3: Input the posture deviation ΔP(t) into the posture evaluation network PEN to obtain the error level score ∈(t); Step S4: Normalize the physiological data and construct the physiological state vector B(t)=[HR * ,GSR * ,EMG * ], calculate the physiological weight coefficient: W(t) = αHR * +βGSR * +γEMG * , where α+β+γ=1; Step S5: ∈(t) and W(t) are jointly input into a feedback function to generate a feedback instruction C(t)=f(ΔP(t), W(t)). Feedback methods include but are not limited to vibration signals, voice prompts, or augmented reality overlays. Step S6: Establish an individual database based on the dancer's historical training data, update the posture evaluation model and feedback control function, and implement a personalized correction strategy.

2. The dance therapy posture correction method based on biofeedback according to claim 1, characterized in that: The posture data acquisition method is completed by combining an inertial measurement unit (IMU) with an image skeleton recognition algorithm, and the Madgwick algorithm is used for posture fusion to improve accuracy.

3. The dance therapy posture correction method based on biofeedback according to claim 1, characterized in that: The posture evaluation network is a temporal model based on the Transformer structure, which is used to identify the posture deviation type and output the corresponding grade score.

4. The dance therapy posture correction method based on biofeedback according to claim 1, characterized in that: The feedback control function satisfies the following form: C(t)=δ1·(t)·V haptic +δ2·(t)·W(t)·S voice Where V haptic Represents the directional vibration signal vector S voice represents the voice prompt sequence, and δ1 and δ2 are weight factors.

5. The dance therapy posture correction method based on biofeedback according to claim 1, characterized in that: The normalization processing method of the physiological state vector is interval normalization, which is in the form of: Where T S is the switching cycle.

6. The dance therapy posture correction method based on biofeedback according to claim 1, characterized in that: When the physiological state weight W(t) exceeds a set threshold, the system automatically reduces the feedback intensity and frequency to reduce the dancer's psychological burden.

7. The dance therapy posture correction method based on biofeedback according to claim 1, characterized in that: The individual database H contains a set of historical data during the training process:

8. The dance therapy posture correction method based on biofeedback according to claim 7, characterized in that: The individual model uses a temporal convolutional network (TCN) structure to predict the optimal feedback strategy at the next moment:

9. The dance therapy posture correction method based on biofeedback according to claim 1, characterized in that: The feedback device includes a wearable vibration module, a voice playback terminal and an AR overlay system, which is used to realize multimodal feedback interaction.

10. The dance therapy posture correction method based on biofeedback according to claim 1, characterized in that: The method is applicable to treatment scenarios including but not limited to the following: rehabilitation of autism spectrum disorders, rehabilitation of Parkinson's disease, posture training for stroke patients, balance improvement for the elderly, basic dance teaching, etc.

Images