Personalized motion posture correction system and method based on multi-modal biomechanical feedback

Abstract

The invention discloses a personalized motion posture correction system and method based on multi-modal biomechanical feedback, and relates to the technical field of emotion recognition and intelligent feedback control. The system comprises the following components: a multi-modal biomechanical data acquisition module, an emotional state sensing module, a posture deviation analysis module, a dynamic feedback decision module and a multi-channel feedback output module. The multi-mode biomechanical data acquisition module acquires biomechanical parameters such as muscle electric signals, joint motion trails and gravity center distribution of a user in real time, and combines emotional related physiological signals such as heart rate variability and skin conductivity acquired by the emotional state sensing module and results of the motion scene recognition unit. According to the method, the personalized biomechanical baseline model is constructed, and the motion posture deviation type and degree of the user can be identified more accurately in a multi-modal data fusion and personalized modeling mode, so that targeted feedback and correction are provided.

Description

Technical Field

[0001] This invention relates to the field of emotion recognition and intelligent feedback control technology, specifically to a personalized motion posture correction system and method based on multimodal biomechanical feedback. Background Technology

[0002] With the increasing awareness of fitness among the general public and the popularization of sports, how to conduct sports training scientifically and effectively, prevent sports injuries, and improve sports performance has become a focus of attention for many sports enthusiasts and professional athletes. As a key factor affecting the effectiveness and safety of sports, the accuracy and standardization of sports posture are directly related to the risk of sports injuries and the improvement of sports efficiency.

[0003] Traditional methods of correcting exercise postures primarily rely on coaches' visual observations or simple video analysis. These methods have limitations such as high subjectivity, poor real-time performance, and inability to quantify assessments. Specifically, coaches' visual observations are easily influenced by factors such as experience and distraction, leading to inconsistent and erroneous assessment results. While video analysis can provide some objective data, it often lacks monitoring of the exerciser's internal physiological and emotional states, making it difficult to achieve truly personalized correction. Furthermore, traditional methods often fail to adjust correction strategies in a timely manner when faced with complex and changing exercise scenarios, affecting the correction effect and exercise experience. More seriously, long-term training with incorrect postures not only reduces exercise efficiency but may also lead to chronic sports injuries, posing a potential threat to the exerciser's health.

[0004] Given the limitations of traditional motion posture correction methods, it is therefore of great importance to develop a personalized motion posture correction system and method based on multimodal biomechanical feedback. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a personalized motion posture correction system and method based on multimodal biomechanical feedback. By integrating advanced sensor technology, biomechanical analysis, emotion recognition, and intelligent feedback control technology, it can achieve comprehensive monitoring and accurate analysis of the athlete's biomechanical characteristics, emotional state, and exercise environment. Compared with traditional methods, this system has higher accuracy, real-time performance, and personalization. It can dynamically adjust the correction strategy according to the athlete's real-time state, effectively reducing the risk of sports injuries and improving athletic performance.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On the one hand, a personalized motion posture correction system based on multimodal biomechanical feedback, the system comprising the following components: a multimodal biomechanical data acquisition module, an emotional state perception module, a posture deviation analysis module, a dynamic feedback decision module, and a multi-channel feedback output module;

[0007] Multimodal biomechanical data acquisition module: Composed of wearable sensor array, used to collect biomechanical parameters such as muscle electrical signals, joint movement trajectory, and center of gravity distribution during the user's movement in real time, and supports multimodal data fusion processing and the construction and dynamic correction of personalized biomechanical baseline models;

[0008] Emotional state perception module: integrates physiological signal sensors and motion scene recognition unit, used to collect emotion-related physiological signals such as heart rate variability and skin conductivity, and identify the user's real-time emotional state by combining motion scene classification results. The motion scene recognition unit adopts a scene classification algorithm with multi-feature fusion.

[0009] Posture Deviation Analysis Module: Based on multimodal biomechanical data, this module uses a deviation calculation model with dynamic threshold correction to identify the type and degree of user motion posture deviation.

[0010] Dynamic feedback decision-making module: Constructs an emotion-feedback adaptation model, and outputs targeted feedback control commands with dynamically adjustable feedback intensity and frequency based on posture deviation analysis results, emotional state perception results, and scene adaptation coefficients.

[0011] Multi-channel feedback output module: includes tactile, auditory, and visual feedback units, which output corresponding feedback signals according to feedback control commands.

[0012] Furthermore, the multimodal biomechanical data acquisition module employs an adaptive dynamic weighting method for data fusion. The specific steps are as follows: First, the raw data for each modality is preprocessed, and motion noise is removed using wavelet transform. Second, initial weights are calculated based on the real-time signal-to-noise ratio and data integrity of each modality, with the initial weights set to 0.35 for electromyography signals, 0.4 for inertial measurement signals, and 0.25 for plantar pressure. Then, the weights are dynamically adjusted according to the stability of each modality during movement. When the signal-to-noise ratio of a certain modality is below -15dB, its weight is reduced by 20%, while the weights of other highly stable modalities are proportionally increased. Finally, the fused biomechanical feature vector is obtained through weighted summation, achieving reliable fusion of multi-source data.

[0013] Furthermore, when constructing the personalized biomechanical baseline model, user bodily function parameters are incorporated for dynamic correction. The baseline correction formula is: B′ j =B j ·(1+δ·P j), where B′ j For the revised personalized baseline, B j The initial baseline was calculated based on the average of three standard movement data collected from the user before exercise. δ = 0.02 is the correction coefficient, determined through baseline adaptation experiments with over 200 users of different physical abilities. P j For the body function parameters corresponding to the j-th monitoring point, the knee baseline correction incorporates the quadriceps muscle strength value, measured using an isokinetic muscle strength testing device at an angular velocity of 60° / s; the lumbar baseline correction incorporates the core muscle endurance value, calculated through a plank test duration. P j The system is recalibrated every 30 days. During the calibration process, users need to complete a basic physical fitness test. The system automatically updates the corrected baseline value to resolve the baseline failure caused by changes in the user's physical condition and ensure the long-term accuracy of posture deviation recognition.

[0014] Furthermore, the motion scene recognition unit employs a multi-feature fusion scene classification algorithm, and the scene adaptation coefficient is calculated as follows: Where S is the scene complexity coefficient, ranging from 0 to 1, q is the number of scene features, which by default includes four core features: motion frequency, motion amplitude, environmental noise, and motion intensity, and F... p Let φ be the normalized value of the p-th feature. Min-max normalization maps the original feature values ​​to the 0-1 interval. p The feature weights are determined by training on labeled datasets of 10 typical motion scenarios, with motion frequency φ = 0.4, motion amplitude φ = 0.3, and environmental noise φ = 0.3. The weight allocation is optimized using a random forest algorithm. When the system detects a scene change, it completes the adaptation and update of biomechanical thresholds and feedback strategies within 500ms to ensure the continuity and accuracy of cross-scene correction and adapt to the personalized needs of different motion scenarios.

[0015] Furthermore, the emotion state perception module adopts an emotion classification model based on spatiotemporal feature fusion. The formula for calculating the emotion state quantification index is: E=γ1·LF / HF+γ2·ΔSC+γ3·S, where E is the quantification value of the emotion state, LF / HF is the low-frequency / high-frequency power ratio of heart rate variability, reflecting the activity of the sympathetic nervous system, with a resting state baseline value of 1.5, calculated by continuously collecting data for 5 seconds using a heart rate variability sensor, ΔSC is the skin conductivity change rate, S is the scene complexity coefficient, and γ1=0.4, γ2=0.35, and γ3=0.25 are weight coefficients, determined through training on an emotion-physiological signal association dataset of 800+ users. The statistical significance of the weights is ensured by a two-sample t-test with P<0.01. This formula achieves deep coupling analysis of physiological signals and scene information, and the emotion recognition accuracy can reach over 92%, which is 18% higher than that of a single physiological signal recognition algorithm.

[0016] Furthermore, the attitude deviation analysis module employs a deviation calculation model with dynamic threshold correction, the formula of which is: Where D is the total posture deviation, m is the number of joints / muscles monitored, and X... j Let B′ be the real-time biomechanical parameter of the j-th monitoring point. j For the corrected personalized baseline value, T j ω is the safety threshold for this parameter. j The weights are determined based on a sports injury risk assessment model: knee joint ω = 0.3, lumbar spine ω = 0.25, ankle joint ω = 0.2, hip joint ω = 0.15, and other joints ω = 0.1. The higher the risk, the greater the weight. This formula solves the problem of poor adaptability of traditional fixed thresholds to users with different physical conditions by combining personalized baselines and dynamic thresholds. The misjudgment rate of posture deviation recognition is reduced to below 8%.

[0017] Furthermore, in the emotion-feedback adaptation model of the dynamic feedback decision module, the feedback intensity adjustment formula is: I=I0·(k1·E′+k2·D′+k3·C), where I is the final feedback intensity, I0 is the base intensity, E′ is the emotion normalization value, D′ is the deviation normalization value, and C is the scene adaptation coefficient, C=0.8 for quiet indoor scenes and C=1.2 for noisy outdoor scenes, determined based on environmental noise sensor data; k1=-0.2 negative weight indicates that the more irritable the emotion, the lower the feedback intensity; k2=0.5 the greater the deviation, the higher the feedback intensity; k3=0.3 satisfies k1+k2+k3=1. The weights are iteratively optimized through user feedback data and updated once every 100 effective corrections. This formula realizes the coordinated control of emotion, deviation degree and scene, and the feedback intensity adjustment response time is ≤100ms, ensuring the real-time performance and adaptability of the feedback.

[0018] Furthermore, the multi-channel feedback output module adopts a three-level frequency modulation method of basic frequency - state correction - response fine-tuning to adjust the feedback frequency. The specific steps are as follows: First, preset the basic frequency according to the type of movement; second, perform a first-level correction based on the emotional state; third, perform a second-level correction in combination with the posture deviation; fourth, monitor the user's feedback response in real time. If the posture is not adjusted after two consecutive feedbacks, the frequency is increased by 10% every 5 seconds. If the adjustment is completed after one feedback, it returns to the second-level correction frequency, realizing dynamic adaptation and precise control of the frequency.

[0019] Furthermore, the system also includes a model iterative optimization unit that uses a feedback effect grading-parameter progressive adjustment method to update the parameters of the emotion-feedback adaptation model. The specific steps are as follows: First, collect user feedback data, including response time to feedback, the difference between the posture adjustment amplitude and the target deviation, and changes in emotional state after feedback; second, set parameter update trigger conditions, starting the update when 50 valid feedback data points are accumulated or 10 consecutive invalid feedback data points are collected; then, adjust the parameters according to the feedback effect grading, increasing the parameter weight for valid feedback by 5%, keeping the parameter weight for delayed feedback unchanged, and decreasing the parameter weight for invalid feedback by 3%; finally, perform parameter calibration every 7 days, correcting parameter drift by comparing recent and historical correction effects, and ensuring that the model always adapts to the user's movement habits and feedback preferences.

[0020] On the other hand, a personalized motion posture correction method based on multimodal biomechanical feedback is characterized by the following specific steps:

[0021] S1. System initialization: Collect user's basic physiological parameters and motion goals, and construct a personalized biomechanical baseline model for the user.

[0022] S2. During exercise, the multimodal biomechanical data acquisition module collects muscle electrical signals and joint motion trajectory data in real time, while the emotional state perception module collects heart rate variability, skin conductivity and scene information simultaneously.

[0023] S3, the posture deviation analysis module compares the collected biomechanical data with the personalized baseline model to calculate the posture deviation degree and deviation type; the emotion state perception module outputs the user's real-time emotion state through an emotion classification model.

[0024] S4. The dynamic feedback decision-making module calls the emotion-feedback adaptation model to generate feedback instructions according to the following rules:

[0025] If the emotional state is focused, choose high-frequency, quantitative feedback;

[0026] If the emotional state is irritability, switch to low-frequency, gentle feedback;

[0027] If the emotional state is fatigue, the correction accuracy requirement will be automatically reduced, and safety warning feedback will be given priority.

[0028] S5. The multi-channel feedback output module executes feedback instructions and records the user's response data to the feedback.

[0029] S6. Optimize the emotion-feedback adaptation model in real time based on user response data to achieve continuous iteration of feedback strategies.

[0030] Compared with existing technologies, this personalized motion posture correction system and method based on multimodal biomechanical feedback has the following advantages:

[0031] I. This invention uses a multimodal biomechanical data acquisition module to collect real-time biomechanical parameters such as muscle electrophysiological signals, joint movement trajectories, and center of gravity distribution from users. Combined with emotion-related physiological signals such as heart rate variability and skin conductivity collected by the emotion state perception module, as well as the results of the motion scene recognition unit, a personalized biomechanical baseline model is constructed. This multimodal data fusion and personalized modeling approach can more accurately identify the type and degree of user's motion posture deviation, thereby providing targeted feedback and correction, greatly improving the accuracy and personalization of motion posture correction.

[0032] Second, this invention employs a dynamic feedback decision module. Based on posture deviation analysis results, emotional state perception results, and scene adaptation coefficients, it outputs targeted feedback control commands with dynamically adjustable feedback intensity and frequency. This intelligent feedback mechanism can adjust the feedback strategy according to the user's real-time state. For example, it provides high-frequency, quantitative feedback when the user is focused, and switches to low-frequency, gentle feedback when the user is agitated, thereby enhancing the user experience. At the same time, the system also includes a model iteration optimization unit. By collecting user feedback data and optimizing the emotion-feedback adaptation model parameters in real time, it achieves continuous iteration of the feedback strategy and promotes the continuous optimization of the user's exercise habits.

[0033] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0035] Figure 1 This is a schematic diagram of a personalized motion posture correction system based on multimodal biomechanical feedback.

[0036] Figure 2 This is a flowchart illustrating a personalized motion posture correction method based on multimodal biomechanical feedback. Detailed Implementation

[0037] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0038] Example 1

[0039] The user is a 28-year-old female whose exercise goal is to improve posture issues such as rounded shoulders, hunched back, and forward head posture during yoga practice. After the system is activated, it first collects the user's physical function parameters, including height, weight, shoulder width, cervical spine mobility, and shoulder muscle strength. Combined with a preset initial baseline, a personalized biomechanical baseline model is generated for the user using a baseline correction formula: B′ j =B j ·(1+δ·P j ), where B′ j For the revised personalized baseline, B j The initial baseline is given, δ = 0.02 is the correction factor, and P... j The system sets the physical function parameters corresponding to the j-th monitoring point to ensure that the baseline matches the user's physical function. At the same time, the system presets the basic feedback intensity and basic frequency corresponding to yoga exercises to complete the initial configuration.

[0040] When users practice yoga poses such as Mountain Pose and Downward-Facing Dog, the multimodal biomechanical data acquisition module uses a wearable sensor array to collect real-time biomechanical parameters such as electromyography (EMG) signals of the shoulder and neck muscles, the motion trajectory of the cervical spine and shoulder joints, and plantar pressure distribution. In the data preprocessing stage, wavelet transform is used to remove noise generated during movement. Data fusion employs an adaptive dynamic weighting method, with initial weights set at 0.35 for EMG signals, 0.4 for inertial measurement signals, and 0.25 for plantar pressure. Simultaneously, the weights are dynamically adjusted based on the real-time signal-to-noise ratio (SNR) and integrity of each modality. If the SNR of a certain modality falls below -15dB, its weight is reduced by 2. 0%, with the remaining high-stability modal weights increased proportionally. The emotion state perception module synchronously collects the user's heart rate variability and skin conductivity change rate, and combines the multi-feature fusion scene classification results of the motion scene recognition unit. The user's real-time emotion state is identified through an emotion classification model based on spatiotemporal feature fusion. The formula is: E=γ1·LF / HF+γ2·ΔSC+γ3·S, where E is the quantification value of the emotion state, LF / HF is the low-frequency / high-frequency power ratio of heart rate variability, ΔSC is the skin conductivity change rate, S is the scene complexity coefficient, and γ1=0.4, γ2=0.35, and γ3=0.25 are weight coefficients.

[0041] The posture deviation analysis module compares real-time collected biomechanical parameters with a personalized biomechanical baseline model, and uses a dynamic threshold correction deviation calculation model, the formula of which is: Where D is the total posture deviation, m is the number of joints / muscles monitored, and X... j Let B′ be the real-time biomechanical parameter of the j-th monitoring point. j For the corrected personalized baseline value, T j ω is the safety threshold for this parameter. j Using weights, the total posture deviation and specific deviation types of monitoring points such as the cervical spine and shoulder joint are calculated to clarify the degree of the user's forward head posture deviation. The emotion state perception module uses the emotion state quantification index calculation formula, combined with the low-frequency / high-frequency power ratio of heart rate variability, skin conductivity change rate, and scene adaptation coefficient, to output the user's emotion state quantification value, determining whether the user is currently focused, agitated, or fatigued. Figure 1 As shown.

[0042] The dynamic feedback decision module calls the emotion-feedback adaptation model, combining the results of posture deviation analysis, emotional state perception, and scene adaptation coefficients to generate feedback instructions. If the user's emotional state is focused, the feedback instruction is set to high-frequency, quantitative feedback to ensure accurate and efficient correction. The feedback intensity adjustment formula is: I = I0·(k1·E′+k2·D′+k3·C), where I is the final feedback intensity, I0 is the base intensity, E′ is the emotion normalization value, D′ is the deviation normalization value, and C is the scene adaptation coefficient. If the user becomes irritable due to the difficulty of the posture, the feedback instruction is switched to low-frequency, gentle feedback to avoid aggravating the user's negative emotions. If the user becomes fatigued due to prolonged practice, the system automatically reduces the correction accuracy requirement and prioritizes outputting safety warning feedback to remind the user to avoid excessive muscle strain. The feedback intensity is calculated using the feedback intensity adjustment formula, with the base intensity preset based on the characteristics of yoga exercise, and the weighting coefficients set to k1 = -0.2, k2 = 0.5, and k3 = 0.3.

[0043] The multi-channel feedback output module executes feedback commands. The tactile feedback unit generates vibration prompts at corresponding shoulder and neck positions via wearable devices. The auditory feedback unit outputs voice prompts such as "shoulders should be extended backward" and "neck should remain neutral" via Bluetooth headphones. The visual feedback unit displays a real-time diagram of shoulder and neck posture deviations on a screen next to the yoga mat. Simultaneously, the system records the user's response time to feedback, the difference between the posture adjustment range and the target deviation, and changes in emotional state after feedback, providing a basis for model optimization. The feedback frequency is adjusted using a three-level frequency tuning method: base frequency - state correction - response fine-tuning. First, a base frequency is preset according to the yoga exercise type. Then, a first-level correction is performed based on emotional state, followed by a second-level correction based on posture deviation. If the user's posture does not adjust after two consecutive feedbacks, the feedback frequency is increased by 10% every 5 seconds. If adjustment is completed after one feedback, the frequency returns to the second-level correction frequency. Figure 2 As shown.

[0044] The model iterative optimization unit uses a feedback effect grading-parameter incremental adjustment method to update the emotion-feedback adaptation model parameters. When the system has collected 50 valid feedback data, it initiates parameter updates: the parameter weights corresponding to valid feedback increase by 5%, the parameter weights corresponding to delayed feedback remain unchanged, and the parameter weights corresponding to invalid feedback decrease by 3%. The system performs parameter calibration every 7 days, compares recent and historical correction effects, corrects parameter drift, and ensures that subsequent feedback strategies are more in line with users' practice habits and physical conditions.

[0045] Example 2

[0046] The user is a 35-year-old male whose exercise goal is to correct his knee valgus posture while running and reduce the risk of knee joint injury. During the system initialization phase, the system collects data on the user's physical function parameters and exercise habits, such as leg length, knee circumference, calf muscle strength, and running cadence preference. Based on the initial baseline, a personalized biomechanical baseline model is constructed using a baseline correction formula. The baseline parameters of knee joint-related monitoring points are optimized. At the same time, the basic feedback intensity and basic frequency are preset according to the characteristics of running exercise to complete the system configuration.

[0047] When a user runs in an outdoor park, the multimodal biomechanical data acquisition module collects biomechanical parameters in real time, such as electromyography (EMG) signals of the front / inner thigh muscles, knee joint movement trajectory, and plantar pressure distribution, through sensors embedded in the running shoes and worn on the legs. Data preprocessing uses wavelet transform to remove road vibration noise during outdoor running. Data fusion uses an adaptive dynamic weighting method, with initial weights of 0.35 for EMG signals, 0.4 for inertial measurement signals, and 0.25 for plantar pressure. If the outdoor environment causes the signal-to-noise ratio of a certain modality to be lower than -15dB, the weight of that modality is reduced by 20%, while the weights of other highly stable modalities are increased proportionally. The emotion state perception module collects the user's heart rate variability and skin conductivity change rate. The motion scene recognition unit determines that it is an outdoor noisy scene through a multi-feature fusion scene classification algorithm with a scene adaptation coefficient C = 1.2. Combining the above data, the user's emotional state is identified through an emotion classification model based on spatiotemporal feature fusion.

[0048] The posture deviation analysis module employs a dynamic threshold-corrected deviation calculation model, comparing real-time collected knee joint biomechanical parameters with a personalized baseline model to calculate the total posture deviation of knee valgus, clarifying the severity of the deviation and whether there is a more pronounced deviation on one knee. The emotion state perception module uses an emotion state quantification index calculation formula, integrating the low-frequency / high-frequency power ratio of heart rate variability, skin conductivity change rate, and scene adaptation coefficient to output a quantitative emotion value, determining whether the user is focused on running, experiencing fatigue and irritability, or fatigue due to excessive exercise. Figure 1 As shown.

[0049] The dynamic feedback decision-making module calls the emotion-feedback adaptation model to generate targeted feedback instructions. If the user's emotional state is focused, the feedback instructions are set to high frequency and quantitative to help the user quickly adjust their knee posture. If the user is irritable due to outdoor noise or disrupted running rhythm, the feedback is switched to low frequency and gentle to avoid interfering with the user's running rhythm. If the user experiences fatigue after a long run, the system lowers the correction accuracy requirements and prioritizes outputting safety warning feedback to remind the user to control running speed and distance to avoid excessive load on the knee joint. The feedback intensity is calculated using a feedback intensity adjustment formula, with the base intensity adapted to the running exercise needs. The weighting coefficients are k1 = -0.2, k2 = 0.5, and k3 = 0.3, and the scene adaptation coefficient is set to 1.2 for noisy outdoor scenes.

[0050] The multi-channel feedback output module executes feedback commands: the tactile feedback unit generates vibrations of different intensities on the inner / outer side of the knee via a wearable device on the lower leg; the auditory feedback unit outputs simple voice prompts such as "knee outward" and "keep feet parallel" via sports headphones; and the visual feedback unit displays a real-time diagram of the knee joint posture and deviation prompts on the sports watch screen. The system simultaneously records data such as the user's response time to feedback, the difference between the knee posture adjustment range and the target deviation, and emotional changes after feedback. The feedback frequency is adjusted using a three-level frequency modulation method: first, a preset base frequency for running is used; then, a first-level correction is performed based on emotional state; and a second-level correction is performed based on posture deviation. If the posture is not adjusted after two consecutive feedbacks, the feedback frequency is increased by 10% every 5 seconds; if the posture is adjusted correctly after one feedback, the second-level correction frequency is restored. Figure 2 As shown.

[0051] The model iterative optimization unit updates model parameters according to feedback effect grading and parameter incremental adjustment method. When 50 valid feedback data are collected or 10 consecutive invalid feedback data are found, parameter update is initiated: the parameter weight corresponding to valid feedback increases by 5%, the parameter weight corresponding to delayed feedback remains unchanged, and the parameter weight corresponding to invalid feedback decreases by 3%. The system performs parameter calibration every 7 days, compares recent and historical correction effects, corrects parameter drift, and makes the feedback strategy more in line with the user's outdoor running rhythm and physical endurance, continuously improving the correction effect of knee valgus posture.

[0052] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A personalized motion posture correction system based on multimodal biomechanical feedback, characterized in that, The system comprises the following components: a multimodal biomechanical data acquisition module, an emotion state perception module, a posture deviation analysis module, a dynamic feedback decision-making module, and a multi-channel feedback output module. Multimodal biomechanical data acquisition module: Composed of wearable sensor array, used to collect biomechanical parameters such as muscle electrical signals, joint movement trajectory, and center of gravity distribution during the user's movement in real time, and supports multimodal data fusion processing and the construction and dynamic correction of personalized biomechanical baseline models; Emotional state perception module: integrates physiological signal sensors and motion scene recognition unit, used to collect emotion-related physiological signals such as heart rate variability and skin conductivity, and identify the user's real-time emotional state by combining motion scene classification results. The motion scene recognition unit adopts a scene classification algorithm with multi-feature fusion. Posture Deviation Analysis Module: Based on multimodal biomechanical data, this module uses a deviation calculation model with dynamic threshold correction to identify the type and degree of user motion posture deviation. Dynamic feedback decision-making module: Constructs an emotion-feedback adaptation model, and outputs targeted feedback control commands with dynamically adjustable feedback intensity and frequency based on posture deviation analysis results, emotional state perception results, and scene adaptation coefficients. Multi-channel feedback output module: includes tactile, auditory, and visual feedback units, which output corresponding feedback signals according to feedback control commands.

2. The personalized motion posture correction system based on multimodal biomechanical feedback according to claim 1, characterized in that, The multimodal biomechanical data acquisition module employs an adaptive dynamic weighting method for data fusion. The specific steps are as follows: First, the raw data for each modality is preprocessed, and motion noise is removed using wavelet transform. Second, initial weights are calculated based on the real-time signal-to-noise ratio and data integrity of each modality. The initial weight for electromyography (EMG) signals is set to 0.35, for inertial measurement signals 0.4, and for plantar pressure 0.

25. Then, the weights are dynamically adjusted according to the stability of each modality during movement. When the signal-to-noise ratio of a certain modality is below -15dB, its weight is reduced by 20%, while the weights of other highly stable modalities are proportionally increased. Finally, the fused biomechanical feature vector is obtained through weighted summation.

3. The personalized motion posture correction system based on multimodal biomechanical feedback according to claim 1, characterized in that, When constructing the personalized biomechanical baseline model, user bodily function parameters are incorporated for dynamic correction. The baseline correction formula is: B′ j =B j ·(1+δ·P j ), where B j ′ represents the corrected personalized baseline, B j The initial baseline is given, δ = 0.02 is the correction factor, and P... j Let be the body function parameters corresponding to the j-th monitoring point.

4. The personalized motion posture correction system based on multimodal biomechanical feedback according to claim 1, characterized in that, The motion scene recognition unit employs a multi-feature fusion scene classification algorithm, and the scene adaptation coefficient is calculated as follows: Where S is the scene complexity coefficient, q is the number of scene features, and F p Let φ be the normalized value of the p-th feature. p These are the feature weights.

5. A personalized motion posture correction system based on multimodal biomechanical feedback according to claim 1, characterized in that, The emotion state perception module adopts an emotion classification model based on spatiotemporal feature fusion. The formula for calculating the emotion state quantification index is: E=γ1·LF / HF+γ2·ΔSC+γ3·S, where E is the emotion state quantification value, LF / HF is the low-frequency / high-frequency power ratio of heart rate variability, ΔSC is the skin conductivity change rate, S is the scene complexity coefficient, and γ1=0.4, γ2=0.35, and γ3=0.25 are weighting coefficients.

6. The personalized motion posture correction system based on multimodal biomechanical feedback according to claim 1, characterized in that, The attitude deviation analysis module uses a dynamic threshold correction deviation calculation model, the formula of which is: Where D is the total posture deviation, m is the number of joints / muscles monitored, and X... j Let B be the real-time biomechanical parameter of the j-th monitoring point. j ′ represents the corrected personalized baseline value, T j ω is the safety threshold for this parameter. j As weight.

7. A personalized motion posture correction system based on multimodal biomechanical feedback according to claim 1, characterized in that, In the emotion-feedback adaptation model of the dynamic feedback decision module, the feedback intensity adjustment formula is: I=I0·(k1·E′+k2·D′+k3·C), where I is the final feedback intensity, I0 is the basic intensity, E′ is the emotion normalization value, D′ is the deviation normalization value, C is the scene adaptation coefficient, C=0.8 for quiet indoor scenes and C=1.2 for noisy outdoor scenes, determined based on environmental noise sensor data; k1=-0.2, k2=0.5, k3=0.3, satisfying k1+k2+k3=1.

8. A personalized motion posture correction system based on multimodal biomechanical feedback according to claim 1, characterized in that, The multi-channel feedback output module uses a three-level frequency modulation method of basic frequency - state correction - response fine-tuning to adjust the feedback frequency. The specific steps are as follows: First, preset the basic frequency according to the type of movement; second, perform first-level correction based on emotional state; third, perform second-level correction in combination with posture deviation; fourth, monitor the user's feedback response in real time. If the posture is not adjusted after two consecutive feedbacks, the frequency is increased by 10% every 5 seconds. If the adjustment is completed after one feedback, it returns to the second-level correction frequency.

9. A personalized motion posture correction system based on multimodal biomechanical feedback according to claim 1, characterized in that, The system also includes a model iterative optimization unit that uses a feedback effect grading-parameter incremental adjustment method to update the parameters of the emotion-feedback adaptation model. The specific steps are as follows: First, collect user feedback data, including the response time to feedback, the difference between the posture adjustment amplitude and the target deviation, and the change in emotional state after feedback; second, set parameter update trigger conditions, and start the update when 50 valid feedback data or 10 consecutive invalid feedback data are accumulated; then, adjust the parameters according to the feedback effect grading, with the parameter weight corresponding to valid feedback increasing by 5%, the parameter weight corresponding to delayed feedback remaining unchanged, and the parameter weight corresponding to invalid feedback decreasing by 3%; finally, perform parameter calibration every 7 days, and correct parameter drift by comparing recent and historical correction effects.

10. A personalized motion posture correction method based on multimodal biomechanical feedback, applicable to the personalized motion posture correction system based on multimodal biomechanical feedback as described in any one of claims 1-9, characterized in that, The specific steps of this method are as follows: S1. System initialization: Collect user's basic physiological parameters and motion goals, and construct a personalized biomechanical baseline model for the user. S2. During exercise, the multimodal biomechanical data acquisition module collects muscle electrical signals and joint motion trajectory data in real time, while the emotional state perception module collects heart rate variability, skin conductivity and scene information simultaneously. S3, the posture deviation analysis module compares the collected biomechanical data with the personalized baseline model to calculate the posture deviation degree and deviation type; the emotion state perception module outputs the user's real-time emotion state through an emotion classification model. S4. The dynamic feedback decision-making module calls the emotion-feedback adaptation model to generate feedback instructions according to the following rules: If the emotional state is focused, choose high-frequency, quantitative feedback; If the emotional state is irritability, switch to low-frequency, gentle feedback; If the emotional state is fatigue, the correction accuracy requirement will be automatically reduced, and safety warning feedback will be given priority. S5. The multi-channel feedback output module executes feedback instructions and records the user's response data to the feedback. S6. Optimize the emotion-feedback adaptation model in real time based on user response data to achieve continuous iteration of feedback strategies.

Images