A real-time feedback interaction guiding method and system for martial arts

Abstract

This application provides a real-time feedback interactive guidance method and system for martial arts, relating to the field of sports training technology. The key technical points are: acquiring real-time motion data and support status data of the user during exercise; determining the user's body tilting torque and the corrective torque generated by specific limb segments based on this data; and determining whether the movement of a specific limb segment is a physiological compensatory movement or a technical error based on the direction and amplitude relationship between the two. The real-time feedback interactive guidance method and system for martial arts provided by this application has the advantages of effectively distinguishing between physiological compensatory movements and technical errors, avoiding misjudgment and negative intervention of physiological compensation, thereby improving the safety, effectiveness, and user experience of training.

Description

Technical Field

[0001] This application relates to the field of sports training technology, and more specifically, to a method and system for real-time feedback and interactive guidance in martial arts. Background Technology

[0002] With the development of intelligent sports training, sports such as martial arts, which require high levels of body control and dynamic balance, are increasingly relying on multi-sensor systems for real-time motion acquisition and feedback guidance. Existing systems typically combine optical vision devices, wearable inertial measurement units, and ground pressure sensors to acquire key skeletal points, joint posture / angular velocity, inertial information, and plantar pressure distribution, and output voice, tactile, or visual correction prompts based on preset motion templates or threshold rules.

[0003] Existing judgment units typically employ geometric template-based comparison and fixed threshold rules. That is, they directly compare the current joint angle and center of gravity trajectory with a preset template; exceeding the threshold triggers correction. The feedback output unit includes bone conduction voice and on-screen prompts, and sets a high-priority strategy for immediate voice correction upon exceeding the threshold; the voice can interrupt other prompts.

[0004] In high-dynamic movements such as single-leg support accompanied by rapid rotation, practitioners often find themselves in a state of critical balance. To avoid tipping, the body will exhibit short, rapid stability recovery movements, such as arm swings, trunk twists, or minor step adjustments. These movements, in a mechanical sense, serve to generate a recovery tendency and pull the center of gravity projection back to the support area. Because current judgment processes primarily rely on whether the instantaneous geometric position / angle conforms to the template, lacking recognition of the stability state and its dynamic contribution, if compensatory movements cause the limb posture to deviate from the template, it may be judged as a technical error and trigger high-priority voice correction. If this correction arrives within the compensation movement execution window, the external correction and internal balance recovery signals conflict in the control direction, potentially leading to suppressed compensation, sluggish movements, or stiffness. This results in failure to recover the center of gravity and instability, creating a counterproductive effect where the more correction is applied, the more dangerous it becomes.

[0005] Furthermore, under high dynamic conditions, visual tracking, inertial measurement, and pressure sampling may experience short-term inconsistencies or errors. This is not an independent problem, but will further increase the probability that compensation is misjudged as an error, making the aforementioned conflict corrections more likely to be triggered.

[0006] It is evident that the core contradiction of existing technologies lies in the fact that when an action enters the dynamic limit region, the system still uses template conformity as the immediate correction trigger condition and adopts rigid high-priority feedback, lacking a mechanism to identify and suppress conflicting corrections of actions that deviate in order to maintain balance.

[0007] To address the aforementioned problems, this application proposes a new solution. Summary of the Invention

[0008] The purpose of this application is to provide a real-time feedback interactive guidance method and system for martial arts, which has the advantages of effectively distinguishing between physiological compensatory movements and technical errors, avoiding misjudgment and negative intervention of physiological compensation, thereby improving the safety, effectiveness and user experience of training.

[0009] This application provides a real-time feedback and interactive guidance method for martial arts exercises, the technical solution of which is as follows: include: Acquire real-time motion data and support status data of the user during exercise; Based on real-time motion data and support status data, the tilting moment of the user's body is determined. The tilting moment is used to characterize the instability trend of the user's body relative to the support area. Based on real-time motion data, the correction torque generated by specific limb segments of the user is determined. The correction torque is used to characterize the contribution of the motion of specific limb segments of the user to the recovery of body posture. Based on the directional and amplitude relationships between the corrective torque and the tilting torque, it can be determined whether the movement of a specific limb segment is a physiological compensatory movement or a technical error movement. The output strategy of the control feedback output device is determined based on the judgment result. When a physiological compensatory action is identified, a feedback inhibition strategy targeting a specific limb segment is executed to block the output of immediate corrective instructions. When a technical error is detected, a feedback strategy targeting specific limb segments is executed to output immediate corrective instructions.

[0010] The above approach can distinguish between physiological compensatory movements and technical errors, avoid misjudging and negatively interfering with physiological compensation, and improve the safety, effectiveness, and user experience of training.

[0011] Furthermore, this application also proposes a method for determining whether a movement of a specific limb segment is a physiological compensatory movement or a technical error movement based on the directional and amplitude relationship between the corrective torque and the tipping torque, including: Compare the direction of the corrective torque with the direction of the overturning torque, and calculate the first difference between the magnitude of the corrective torque and the preset effective compensation threshold. When the direction of the corrective torque is opposite to the direction of the tilting torque, and the first difference is greater than zero, the movement of a specific limb segment is determined to be a physiological compensatory movement. When the direction of the corrective torque is the same as the direction of the tilting torque, or when the first difference is less than or equal to zero, the movement of a specific limb segment is determined to be a technical error.

[0012] The above scheme provides a specific logic for determining whether a physiological compensatory action or a technical error action, making the distinction more accurate and operable.

[0013] Furthermore, this application also proposes that the steps for determining the user's tilting torque based on real-time motion data and support status data include: The position of the user's total center of mass horizontally projected onto the ground plane is determined based on real-time motion data. The boundaries of the ground support area for the user's body are determined based on the support status data; Calculate the shortest horizontal distance from the horizontal projection point to the boundary of the ground support area, and determine the deviation direction corresponding to the shortest horizontal distance; Calculate the tipping moment based on the user's total body mass and gravitational acceleration.

[0014] The above scheme provides a detailed method for calculating the tipping moment, enhancing the accuracy of tipping trend assessment.

[0015] Furthermore, this application also proposes that the method further includes the following steps: The internal clock of the inertial measurement unit that collects real-time motion data is set as the global time reference; For the data streams of the optical vision device that collects real-time motion data and the pressure sensing device that collects support status data, time backtracking interpolation is performed based on the global time reference to align the data streams of the optical vision device and the pressure sensing device to the time axis sampling points of the inertial measurement unit. The aligned data stream is filtered, and the length of the filtering window is adjusted according to the noise variance of the current signal.

[0016] The above solution solves the problems of time synchronization and noise processing of multi-sensor data, and improves the accuracy and robustness of data fusion.

[0017] Furthermore, this application also proposes a step of filtering the aligned data stream, wherein the filter window length of the filtering process is adjusted according to the noise variance of the current signal, including: The noise estimation window is formed by selecting the N most recent sampled values ​​of the current signal within the sliding circular buffer. The noise variance estimate σ² is calculated for the sampled values ​​within the noise estimation window. The noise variance estimate σ² is determined by the average of the squared deviations of the sampled values ​​relative to the mean within the estimation window, or by the variance of adjacent sampled difference sequences. The noise variance estimate σ² is input into a preset monotonic mapping function f(·) to obtain the target filtering window length L-target, such that L-target increases when σ² increases and decreases when σ² decreases, and L-target is restricted between the preset minimum window length L-min and maximum window length L-max. When |L-target−L-curr| is less than the preset change threshold ΔL, the current window length L-curr remains unchanged, or the window length is updated only when L-target satisfies the same increasing / decreasing trend within M consecutive noise estimation windows; When the update condition is met, the current filter window length is updated to L-new, and filtering is performed based on the updated window length.

[0018] The above scheme provides an adaptive filtering window length adjustment mechanism, which can dynamically optimize the filtering effect based on signal noise and further improve data quality.

[0019] Furthermore, this application also proposes that the steps for determining the user's tilting torque based on real-time motion data and support status data include: Real-time monitoring of the user's torso angular velocity; When the trunk angular velocity is lower than the preset dynamic threshold, a preset standard weight allocation strategy is used to fuse real-time motion data and support status data to calculate the body's center of mass position. When the torso angular velocity exceeds the dynamic threshold, the trust weight of visual data in real-time motion data is reduced, while the trust weight of inertial data and support status data in real-time motion data is increased. The body center of mass position is calculated based on the adjusted trust weight, and the tipping torque is calculated based on the body center of mass position.

[0020] The above scheme introduces a dynamic weight adjustment mechanism based on torso angular velocity, which effectively addresses the sensor reliability inversion problem under high dynamic conditions and improves the accuracy of centroid position calculation.

[0021] Furthermore, this application also proposes a step of calculating the body's center of mass position based on the adjusted trust weights, and calculating the tipping moment based on the body's center of mass position, including: Obtain the horizontal projection position of the body's center of mass P-com-vis calculated based on visual data, and the horizontal projection position of the body's center of mass P-com-imu calculated based on inertial data; Obtain the adjusted visual trust weight w-vis and inertial trust weight w-imu, and normalize the trust weights to satisfy: w-vis + w-imu = 1; The centroid horizontal projection positions are weighted and fused according to the normalized trust weights to obtain the fused centroid horizontal projection position P-com. The weighted fusion satisfies: P-com = w-vis P-com-vis + w-imu P-com-imu; The boundary of the ground support area is determined based on the support status data, and the shortest horizontal distance d from the horizontal projection position P-com of the fused centroid to the boundary of the ground support area is calculated to determine the horizontal deviation direction of the horizontal projection position of the fused centroid relative to the boundary of the ground support area. The magnitude of the tipping moment, abs(M-tip), is calculated based on the user's total body mass m-total and gravitational acceleration g. The magnitude of the tipping moment satisfies the following: abs(M-tip) = m-total g d; And make the direction of the tilting moment consistent with the direction of the horizontal deviation.

[0022] The above scheme provides detailed formulas and steps for weighted fusion of center of mass position and calculation of tipping moment, further improving the accuracy of tipping risk assessment under high dynamic conditions.

[0023] Furthermore, this application also proposes that the method further includes the following steps: Calculate the absolute horizontal velocity vector of key foot points in real-time motion data, and the relative horizontal velocity vector of the pressure center in support state data; Calculate the velocity divergence between the absolute horizontal velocity vector and the relative horizontal velocity vector; Calculate the correlation coefficient between velocity divergence and trunk acceleration signals in real-time motion data; When the velocity divergence exceeds the preset divergence threshold and the correlation coefficient is lower than the preset active motion threshold, a passive slip event is determined to have occurred, and the output of the immediate correction command is blocked.

[0024] The above scheme introduces a passive slippage event detection mechanism, which effectively avoids misjudgment and incorrect feedback caused by sensor pad slippage.

[0025] Furthermore, this application also proposes that the method further includes the following steps: Monitor real-time motion data and support status data to identify dynamic time windows where all speed indicators are below the static threshold; Within a dynamic time window, the coordinates of multiple key points on the foot and the coordinates of the pressure center are collected as paired samples. Solve the two-dimensional rigid transformation parameters, including translation and rotation, based on paired samples; The coordinate transformation matrix, updated based on two-dimensional rigid transformation parameters, is used to map real-time motion data and support state data. This matrix is ​​used to transform the support state data into a coordinate system consistent with the real-time motion data during the process of determining the tilting moment of the user's body based on the real-time motion data and support state data. This process determines the boundary of the support area and calculates the lever arm distance of the body's center of mass projection point relative to the boundary of the support area, thereby calculating the tilting moment.

[0026] The above scheme provides a method for online calibration of the coordinate transformation matrix, which solves the coordinate system mismatch problem caused by the slight physical slippage of the pressure sensing pad and ensures the accuracy of data fusion.

[0027] Furthermore, this application also proposes a real-time feedback and interactive guidance system for martial arts, used to implement the above-mentioned scheme, including: The acquisition module is used to acquire real-time motion data and support status data of the user during exercise; The first determining module determines the tilting moment of the user's body based on real-time motion data and support status data. The tilting moment is used to characterize the instability trend of the user's body relative to the support area. The second determining module determines the correction torque generated by a specific limb segment of the user based on real-time motion data. The correction torque is used to characterize the contribution of the motion of the specific limb segment of the user to the recovery of body posture. The judgment module determines whether the movement of a specific limb segment is a physiological compensatory movement or a technical error movement based on the directional and amplitude relationship between the correction torque and the tilting torque. The control module controls the output strategy of the feedback output device based on the judgment result. When a physiological compensatory action is identified, a feedback inhibition strategy targeting a specific limb segment is executed to block the output of immediate corrective instructions. When a technical error is detected, a feedback strategy targeting specific limb segments is executed to output immediate corrective instructions.

[0028] The above scheme provides a system for implementing the above method, enabling the method to be practically deployed and applied, and demonstrating good practicality.

[0029] As described above, the real-time feedback interactive guidance method and system for martial arts provided in this application acquires real-time motion data and support status data of the user during exercise. Based on this data, it determines the user's body tilting torque and the corrective torque generated by specific limb segments. Based on the direction and amplitude relationship between these two data, it intelligently determines whether the movement of a specific limb segment is a physiological compensatory movement or a technical error. When it is determined to be a physiological compensatory movement, a feedback inhibition strategy is implemented to avoid interfering with the user's instinctive balance recovery; when it is determined to be a technical error, a feedback prompt strategy is implemented, outputting immediate corrective instructions. This effectively distinguishes between physiological compensatory movements and technical errors, avoiding misjudgment and negative intervention of physiological compensation, thereby improving the safety, effectiveness, and user experience of training. Attached Figure Description

[0030] Figure 1 This is a flowchart illustrating a real-time feedback and interactive guidance method for martial arts provided in this application.

[0031] Figure 2 A schematic diagram of the structure of the real-time feedback and interactive guidance system for martial arts provided in this application.

[0032] Figure 3 This application provides a schematic diagram of the application scenario and feedback control logic of a real-time feedback interactive guidance method for martial arts.

[0033] In the diagram: 210, Acquisition module; 220, First determination module; 230, Second determination module; 240, Judgment module; 250, Control module. Detailed Implementation

[0034] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0035] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0036] Please refer to Figures 1 to 3 This application proposes a real-time feedback interactive guidance method for martial arts, including: S110. Obtain real-time motion data and support status data of the user during exercise; S120. Based on real-time motion data and support status data, determine the tilting moment of the user's body. The tilting moment is used to characterize the instability trend of the user's body relative to the support area. S130. Based on real-time motion data, determine the correction torque generated by a specific limb segment of the user. The correction torque is used to characterize the contribution of the motion of the specific limb segment of the user to the recovery of body posture. S140. Based on the directional and amplitude relationship between the corrective torque and the tilting torque, determine whether the movement of a specific limb segment is a physiological compensatory movement or a technical error movement. S150. Control the output strategy of the feedback output device based on the judgment result: When a physiological compensatory action is identified, a feedback inhibition strategy targeting a specific limb segment is executed to block the output of immediate corrective instructions. When a technical error is detected, a feedback strategy targeting specific limb segments is executed to output immediate corrective instructions.

[0037] This application first establishes a multimodal data acquisition foundation capable of acquiring user motion state to obtain real-time motion data and support state data of the user. The real-time motion data aims to capture the dynamic changes of various parts of the user's body in three-dimensional space, while the support state data is used to clarify the interaction relationship between the user and the ground.

[0038] In a specific implementation scenario, this data acquisition process can be accomplished by an integrated sensor. For example, multiple high-speed optical cameras are deployed around the training area to capture reflective markers affixed to key joints of the user's body, thereby reconstructing a three-dimensional skeletal motion model. This model can provide information such as the position, posture, velocity, and acceleration of each limb segment. Simultaneously, miniature inertial measurement units (IMUs) are worn on key limb segments of the user, such as the torso, upper arm, and thigh. These units can output angular velocity and linear acceleration information for the corresponding segments, especially providing supplementary information when data drift may occur due to the high-speed rotation of the optical cameras. Support status data is acquired through pressure-sensing pads laid on the ground. These pads can measure the pressure distribution on the user's feet in real time, thereby calculating the location of the pressure center and the boundaries of the ground support area.

[0039] In another implementation scenario, to improve portability and ease of use, real-time motion data acquisition can employ a motion capture solution entirely composed of wearable inertial measurement units (IMUs). By wearing a sufficient number of IMUs on the major limb segments of the human body and combining them with a human body model, the posture and motion parameters of the entire body can be calculated in real time. While this solution may not be as accurate as optical solutions in terms of absolute position, it has high sensitivity and reliability in capturing dynamic changes in posture, especially angular velocity and angular acceleration, which is sufficient to meet subsequent computational needs.

[0040] Alternatively, markerless visual tracking technology based on depth cameras can be used to directly identify and track the skeletal joints of the human body using one or more depth cameras, without requiring the user to wear any markers or sensors. This approach offers the highest level of convenience, but its accuracy and robustness are relatively low.

[0041] Regardless of the specific combination of data acquisition technologies used, their common goal is to provide two types of key information for subsequent mechanical analysis: first, the kinematic and dynamic parameters of each segment of the body, such as mass, position, velocity, and acceleration, which are the basis for calculating the corrective torque; and second, the support relationship between the body and the ground, including the position of the supporting feet and the geometry of the support area, which are the basis for calculating the tipping torque.

[0042] After acquiring real-time motion and support status data, the user's tilting moment is determined. The tilting moment describes the tendency and severity of the user's body losing balance at a given moment. It can be understood as a virtual moment attempting to push the user over. Its calculation relies on determining the position of the body's total center of mass and the ground support area. The position of the body's total center of mass is calculated by dividing the human body into multiple segments such as the head, torso, upper arm, forearm, thigh, and lower leg, and then performing a weighted average based on the mass proportion of each segment and the center of mass positions of each segment determined through motion data.

[0043] The ground support area is a polygon formed by connecting the boundaries of the areas covered by all sensors that generate effective pressure readings, based on data from the pressure-sensing pad. The closer the user's center of gravity is to the horizontal projection of this support area, or even beyond it, the greater the risk of tipping over. The magnitude of the tipping moment is proportional to the horizontal distance from the center of gravity's horizontal projection to the nearest boundary of the support area, and its direction points in the direction of deviation. Actions that help restore balance should produce a mechanical effect that counteracts this tipping moment.

[0044] Simultaneously, based on real-time motion data, the corrective torque generated by specific limb segments of the user is determined. When a user's body tends to tilt, their limbs, especially the arms, torso, and non-supporting legs far from the center of support, undergo rapid movements, generating inertial forces and torques that affect the overall body posture. The corrective torque is the physical quantity used to quantify this effect. For example, when the body tilts to the left, if the user quickly swings their right arm to the right, this action generates a torque pointing to the right; this torque is the corrective torque. Its calculation involves the mass of the limb segment, the lever arm of its center of mass relative to the body's support point, and its acceleration. Through real-time motion data, these parameters can be calculated, thus obtaining the magnitude and direction of the corrective torque generated by a specific limb segment at each moment.

[0045] Based on the directional and amplitude relationships between the corrective torque and the tipping torque, it can be determined whether a movement of a specific limb segment is a physiological compensatory movement or a technical error. If the corrective torque generated by a limb movement is in the opposite direction to the body's tipping torque, this means that the movement is pulling the body back from a tipping state, and therefore the movement has a balance-restoring function. Conversely, if the corrective torque is in the same direction as the tipping torque, then the movement will exacerbate instability and is an error.

[0046] Simply reversing the direction is not enough; the amplitude relationship must also be considered. Only when the amplitude of the opposing corrective torque is large enough to effectively counteract the tilting force can it be considered a successful physiological compensation. This "large enough" standard can be defined by a preset threshold.

[0047] Finally, based on the judgment results, a differentiated feedback output strategy is implemented. When a limb movement is determined to be physiological compensation, a feedback inhibition strategy is implemented to block the output of immediate corrective instructions. This means that at critical moments when the user is instinctively trying to save themselves, the guidance system will choose not to issue prompts. The purpose is to provide the user with an undisturbed control environment, enabling them to better complete the balance recovery task and avoid external instructions causing influence and interference.

[0048] When a movement is identified as a technical error, a feedback strategy is implemented, outputting immediate corrective instructions. Since the movement is confirmed to be detrimental to balance, the feedback can be targeted verbal cues, such as tightening the core or stabilizing the torso, or visual or tactile cues. This strategy ensures that the training's error-correction function operates normally when there is no balance risk conflict.

[0049] Through the above steps, the method proposed in this application dynamically evaluates each limb movement that deviates from the standard movement by calculating and comparing the tilting torque and the correction torque in real time. This achieves intelligent switching of feedback strategies, protects the user's physiological balance mechanism by suppressing feedback at critical moments, and corrects technical errors by providing precise prompts under normal conditions. Thus, while ensuring training safety, it greatly improves the effectiveness and pertinence of guidance.

[0050] Specifically, based on the directional and amplitude relationships between the corrective torque and the tipping torque, the steps to determine whether a movement of a specific limb segment is a physiological compensatory movement or a technical error include: Compare the direction of the corrective torque with the direction of the overturning torque, and calculate the first difference between the magnitude of the corrective torque and the preset effective compensation threshold. When the direction of the corrective torque is opposite to the direction of the tilting torque, and the first difference is greater than zero, the movement of a specific limb segment is determined to be a physiological compensatory movement. When the direction of the corrective torque is the same as the direction of the tilting torque, or when the first difference is less than or equal to zero, the movement of a specific limb segment is determined to be a technical error.

[0051] In three-dimensional space, torque is a vector with magnitude and direction. The directions of two torque vectors can be compared by calculating their dot product. If the dot product is negative, the angle between the two vectors is greater than 90 degrees, indicating opposite directions. If the dot product is positive, the angle is less than 90 degrees, indicating the same direction. If the dot product is zero, the directions are perpendicular. In practical applications, we usually focus on the torque component on the horizontal plane because the tilting of a person primarily occurs in the horizontal direction.

[0052] Calculate the first difference between the magnitude of the corrective torque and a preset effective compensation threshold. This effective compensation threshold is not a fixed value, but a dynamic value related to the current tipping moment. For example, the effective compensation threshold can be set to 60% of the current tipping moment amplitude. This means that the compensation is considered effective only when the magnitude of the corrective torque exceeds 60% of the tipping moment amplitude. Calculate the first difference, which is the magnitude of the corrective torque minus the effective compensation threshold. If this difference is greater than zero, it indicates that the magnitude of the corrective torque meets the validity requirement.

[0053] The final judgment rule is a combination of these two conditions. The limb movement is judged as physiological compensation if and only if the direction of the corrective torque is opposite to the direction of the tilting torque and the first difference is greater than zero. Subsequent feedback inhibition can only be triggered if both the correct direction and sufficient force are met simultaneously.

[0054] Any situation that does not meet the above strict conditions will be classified as a technical error. This includes two main scenarios: The first scenario is where the direction of the corrective torque is the same as the direction of the overturning torque. This is undoubtedly incorrect because it exacerbates instability. The second scenario is where, although the directions are opposite, the first difference is less than or equal to zero, meaning the amplitude of the corrective torque is insufficient to reach the threshold for effective compensation. While this situation does not exacerbate instability, its contribution to restoring balance is minimal. It may be an unintentional, ineffective swaying motion, or a technical error due to insufficient amplitude, and it also needs to be pointed out and corrected.

[0055] Furthermore, the steps for determining the user's tilting torque based on real-time motion data and support status data include: The position of the user's total center of mass horizontally projected onto the ground plane is determined based on real-time motion data. The boundaries of the ground support area for the user's body are determined based on the support status data; Calculate the shortest horizontal distance from the horizontal projection point to the boundary of the ground support area, and determine the deviation direction corresponding to the shortest horizontal distance; Calculate the tipping moment based on the user's total body mass and gravitational acceleration.

[0056] First, the human body model is simplified into several rigidly connected segments, such as the head, neck, torso, upper arm, forearm, hand, thigh, lower leg, and foot. Based on standard anthropometry data, the proportion of each segment's mass to the total body mass, as well as the relative position of each segment's center of mass within its geometry, can be pre-determined. During movement, real-time motion data allows determination of each segment's position and orientation in the global coordinate system. From this, the three-dimensional coordinates of each segment's center of mass can be calculated. Finally, by weighting the coordinates of all segments according to their mass proportions, the total three-dimensional coordinates of the body's center of mass are obtained. Projecting this coordinate vertically onto a horizontal surface yields the horizontal projection point of the total center of mass.

[0057] When a user stands on one foot, the support area is the outline of the area where that foot contacts the ground. When the user stands on both feet, the support area is the smallest convex polygon that encloses the contact area of ​​both feet. This information can be provided by the ground pressure sensing pad. Each sensing unit on the pressure sensing pad can detect the pressure value it receives. The area formed by all sensing units with non-zero pressure values ​​is the total contact area. The boundary of this area can be quickly determined through edge detection.

[0058] After determining the horizontal projection point of the center of mass and the boundary of the ground support area, the shortest horizontal distance from this projection point to the boundary of the support area is calculated. When the projection point is inside the support area, the distance is positive, indicating that the body is in a stable state; the larger the distance, the more stable the body. When the projection point falls exactly on the boundary, the distance is zero, and the body is in a critical equilibrium state. When the projection point exceeds the boundary of the support area, the distance can be defined as negative or its absolute value, indicating that the body has entered an unstable state; the larger the absolute value of the distance, the more severe the instability. Simultaneously, the direction corresponding to this shortest distance also needs to be determined, i.e., the direction from the boundary of the support area to the projection point of the center of mass; this direction is the direction of the body's deviation.

[0059] The physical essence of the tipping moment is the torque generated by total gravity relative to the supporting boundary. Its magnitude is equal to the user's total body mass multiplied by the acceleration due to gravity (i.e., total gravity), and then multiplied by the previously calculated shortest horizontal distance (i.e., the lever arm). The direction of this torque is the same as the direction of deviation. Through this calculation process, the complex body posture and support relationship are transformed into a simple and clear tipping moment vector containing both magnitude and direction, providing a solid physical foundation for subsequent compensation determination.

[0060] In a preferred embodiment, the method further includes the following steps: The internal clock of the inertial measurement unit that collects real-time motion data is set as the global time reference; For the data streams of the optical vision device that collects real-time motion data and the pressure sensing device that collects support status data, time backtracking interpolation is performed based on the global time reference to align the data streams of the optical vision device and the pressure sensing device to the time axis sampling points of the inertial measurement unit. The aligned data stream is filtered, and the length of the filtering window is adjusted according to the noise variance of the current signal.

[0061] Since optical vision devices, pressure sensing devices, and inertial measurement units are three different physical devices, their internal clocks, sampling frequencies, and data transmission delays may differ. If data from different moments is used when calculating the torque, such as using the center of mass position from 0.1 seconds ago and the support area at the current moment, the resulting tilting torque will be incorrect, directly leading to deviations in subsequent judgments and feedback.

[0062] To address this issue, this application proposes establishing a unified time scale by setting the internal clock of the inertial measurement unit (IMU) as the global time reference. This is because IMUs typically have the highest sampling frequency and the lowest latency, resulting in the most accurate timestamps, making them suitable as alignment targets.

[0063] Next, time alignment is performed on the other data streams, and time-backward interpolation is executed on the data streams from the optical vision and pressure sensing devices. Specifically, for each inertial measurement unit's timeline sampling point, the two data points in the vision and pressure data streams whose timestamps are closest to that sampling point are found, one before and one after. Then, using the values ​​of these two data points and their timestamps, linear interpolation or other interpolation algorithms are used to calculate the values ​​that the vision and pressure data should have at that inertial measurement unit sampling point. This process generates multiple data streams that are strictly time-synchronized at each decision point, ensuring that all calculations are based on the same instantaneous body state.

[0064] After data synchronization, filtering is required to improve signal quality. Raw sensor data, especially velocity and acceleration signals obtained through differential calculations, often contain noise. This noise can interfere with torque calculations, causing unnecessary jitter in the results. However, traditional fixed-window-length filters, such as mean filters or Gaussian filters, have a problem: a too-long window provides good filtering but introduces significant delay, making it unsuitable for real-time feedback; a too-short window results in less delay but poorer filtering.

[0065] To address this, this embodiment employs an adaptive filtering strategy. The length of the filtering window is no longer fixed but dynamically adjusted based on the current signal noise level. Specifically, the noise variance of the signal within a short time window is calculated in real time. When an increase in the noise variance is detected, such as when a user is engaged in high-speed, vigorous movement, the length of the filtering window is automatically increased to enhance smoothing and suppress noise. When a decrease in the noise variance is detected, such as when the user is relatively stationary, the length of the filtering window is automatically shortened to reduce latency and improve response speed. This maximizes the quality of the input data while ensuring low latency and high responsiveness.

[0066] Specifically, the steps of filtering the aligned data stream, and adjusting the filter window length based on the noise variance of the current signal, include: The noise estimation window is formed by selecting the N most recent sampled values ​​of the current signal within the sliding circular buffer. The noise variance estimate σ² is calculated for the sampled values ​​within the noise estimation window. The noise variance estimate σ² is determined by the average of the squared deviations of the sampled values ​​relative to the mean within the estimation window, or by the variance of adjacent sampled difference sequences. The noise variance estimate σ² is input into a preset monotonic mapping function f(·) to obtain the target filtering window length L-target, such that L-target increases when σ² increases and decreases when σ² decreases, and L-target is restricted between the preset minimum window length L-min and maximum window length L-max. When |L-target−L-curr| is less than the preset change threshold ΔL, the current window length L-curr remains unchanged, or the window length is updated only when L-target satisfies the same increasing / decreasing trend within M consecutive noise estimation windows; When the update condition is met, the current filter window length is updated to L-new, and filtering is performed based on the updated window length.

[0067] First, to estimate the noise level in real time, a sliding circular buffer is maintained, which stores a series of recent sampled values ​​of the signal. The N most recent sampled values ​​are selected from this buffer to form a noise estimation window. The value of N needs to be chosen to balance the stability of the estimation and the speed of response to changes in noise; for example, it can be set to fifty.

[0068] Next, calculate the noise variance estimate σ² within this window. There are two commonly used methods. The first is to calculate the average of the squared deviations of all sampled values ​​within the window from the mean within the window. This method reflects the degree of signal fluctuation. The second method is to first calculate the differences between adjacent sampled points within the window, forming a difference sequence, and then calculate the variance of this difference sequence. For signals with slow drift, the second method can more accurately reflect the intensity of high-frequency noise.

[0069] After obtaining the noise variance estimate σ², it needs to be mapped to the length of the filter window. This can be achieved using a pre-defined monotonic mapping function f(·). This function must be designed to ensure that as σ² increases, the output target filter window length L-target also increases; and as σ² decreases, L-target also decreases. This can be achieved using a simple linear mapping function. To prevent the window length from becoming too large or too small, it also needs to be limited to a pre-defined reasonable range, namely between the minimum window length L-min and the maximum window length L-max. For example, L-min can be set to three, and L-max can be set to thirty-one.

[0070] Directly using L-target to update the current filter window length L-curr may cause frequent small fluctuations in the window length, resulting in unstable filtering performance. To address this issue, one approach is to set a change threshold ΔL, only updating when the absolute difference between L-target and L-curr exceeds ΔL. Another approach is to add a trend check; only when L-target exhibits the same increasing or decreasing trend across M consecutive noise estimation windows is it considered a genuine change in noise level, and an update is then performed.

[0071] Finally, once the update conditions are met, the current filter window length is officially updated to the new length L-new, and the filtering operation is performed using this new window length starting from the next sampling point. This scheme achieves a balance between filter strength and signal real-time performance.

[0072] In a preferred embodiment, the step of determining the user's tilting torque based on real-time motion data and support status data includes: Real-time monitoring of the user's torso angular velocity; When the trunk angular velocity is lower than the preset dynamic threshold, a preset standard weight allocation strategy is used to fuse real-time motion data and support status data to calculate the body's center of mass position. When the torso angular velocity exceeds the dynamic threshold, the trust weight of visual data in real-time motion data is reduced, while the trust weight of inertial data and support status data in real-time motion data is increased. The body center of mass position is calculated based on the adjusted trust weight, and the tipping torque is calculated based on the body center of mass position.

[0073] In normal, slow-moving motions, optical visual data is generally considered the most reliable source for calculating the body's center of mass position because it provides accurate absolute spatial location information. However, when a user rotates rapidly, optical cameras are prone to motion blur, glare, or limb occlusion, causing the tracked joint points to drift or even be lost. In such cases, the reliability of the visual data drops sharply.

[0074] Conversely, the inertial measurement unit worn by the user and the pressure sensor pad on the ground exhibited better robustness during high-dynamic rotation. The inertial measurement unit accurately captured the rotational angular velocity and attitude changes of the torso, while the pressure sensor pad stably reflected the force applied to the supporting feet. Therefore, at specific moments of high-speed rotation, the reliability of the data sources was reversed: the originally most reliable visual data became the least reliable, while the inertial and pressure data, which were originally auxiliary, became more critical.

[0075] This embodiment is designed to address this situation. By monitoring the user's torso angular velocity in real time, it serves as a simple and effective indicator to determine whether the user is in a highly dynamic rotational state. A preset dynamic threshold is used, for example, 180 degrees per second. When the torso angular velocity is below this threshold, it indicates that the user's movement is relatively gentle. In this case, a standard weighting strategy is employed, assigning a higher trust weight to the visual data to ensure the absolute positional accuracy of the centroid calculation.

[0076] Once the torso angular velocity exceeds the threshold, the system immediately switches to a high-dynamic mode. In this mode, the trust weight of visual data is reduced, for example, from 80% to 20%. Simultaneously, the trust weight of inertial data and support status data is correspondingly increased. This means that when calculating the body's center of mass position, more reliance is placed on the body posture derived from inertial data and the support center determined by pressure data, while less attention is given to potentially inaccurate visual data. This dynamic weight adjustment ensures that even under the most extreme motion conditions, the calculated body center of mass position and subsequent tipping moment remain reliable, thus avoiding misjudgments caused by sensor failure.

[0077] Specifically, the steps of calculating the body's center of mass position based on the adjusted trust weights, and calculating the tipping moment based on that body center of mass position, include: Obtain the horizontal projection position of the body's center of mass P-com-vis calculated based on visual data, and the horizontal projection position of the body's center of mass P-com-imu calculated based on inertial data; Obtain the adjusted visual trust weight w-vis and inertial trust weight w-imu, and normalize the trust weights to satisfy w-vis + w-imu = 1; The centroid horizontal projection positions are weighted and fused according to the normalized trust weights to obtain the fused centroid horizontal projection position P-com. The weighted fusion satisfies P-com = w-vis. P-com-vis + w-imu P-com-imu; The boundary of the ground support area is determined based on the support status data, and the shortest horizontal distance d from the horizontal projection position P-com of the fused centroid to the boundary of the ground support area is calculated to determine the horizontal deviation direction of the horizontal projection position of the fused centroid relative to the boundary of the ground support area. The magnitude of the tipping moment, abs(M-tip), is calculated based on the user's total body mass m-total and gravitational acceleration g. The magnitude of the tipping moment satisfies abs(M-tip) = m-total. g d; And make the direction of the tilting moment consistent with the direction of the horizontal deviation.

[0078] First, the positions of the two centroids are calculated in parallel. One is P-com-vis, which is based on visual data and calculated using the aforementioned segmented centroid method. The other is P-com-imu, which is mainly based on inertial data. For example, the attitude angles measured by the torso inertial measurement unit can be combined with preset human segment lengths and biomechanical models to estimate the overall posture of the body and then calculate the position of the total centroid.

[0079] Then, based on the current torso angular velocity, the adjusted visual trust weight w-vis and inertial trust weight w-imu are obtained. For example, at low speeds, w-vis is 0.8 and w-imu is 0.2; at high speeds, w-vis becomes 0.2 and w-imu becomes 0.8. These two weights are normalized to ensure that their sum is always equal to one.

[0080] Next, a weighted fusion is performed. The final fused centroid horizontal projection position P-com is equal to P-com-vis multiplied by its weight w-vis, plus P-com-imu multiplied by its weight w-imu. This linear weighted fusion method allows the centroid position calculation results to smoothly transition between different modes, avoiding centroid position jumps caused by abrupt weight changes.

[0081] After obtaining the fused centroid position P-com, the subsequent calculation process is the same as described above. That is, the boundary of the ground support area is determined based on the support status data, the shortest horizontal distance d from P-com to the boundary and the direction of deviation are calculated, and finally, the magnitude and direction of the toppling moment are calculated based on the total mass, gravitational acceleration, and distance d.

[0082] The above scheme ensures that the calculation of the tipping moment can utilize the accuracy of vision under normal conditions and the robustness of inertia under high dynamic conditions, thus providing an accurate and reliable instability risk assessment in various complex motion scenarios.

[0083] In a preferred embodiment, the method further includes the following steps: Calculate the absolute horizontal velocity vector of key foot points in real-time motion data, and the relative horizontal velocity vector of the pressure center in support state data; Calculate the velocity divergence between the absolute horizontal velocity vector and the relative horizontal velocity vector; Calculate the correlation coefficient between velocity divergence and trunk acceleration signals in real-time motion data; When the velocity divergence exceeds the preset divergence threshold and the correlation coefficient is lower than the preset active motion threshold, a passive slip event is determined to have occurred, and the output of the immediate correction command is blocked.

[0084] Ideally, the pressure sensor pad should be firmly fixed to the ground. However, in reality, especially during high-intensity martial arts training, the enormous horizontal shear force generated when a practitioner lands can cause the sensor pad to shift or rotate slightly relative to the ground. This physical displacement can lead to a mismatch between the sensor pad's coordinate system and the global visual coordinate system.

[0085] When calculating the tipping moment, the position of the center of mass captured by the vision system needs to be compared with the support area reported by the pressure sensor pad. If the coordinate systems of the two are inconsistent—for example, if the sensor pad has slipped back five centimeters and the coordinate transformation matrix in the system has not been updated—the system will incorrectly assume that the support area is still in its original position. This will cause the calculated distance from the center of mass to the support boundary to be overestimated, thus underestimating the true tipping moment. A posture that is already on the verge of instability may be misjudged as stable. Based on this erroneous judgment, when the practitioner makes a correct physiological compensatory movement, the system may instead identify it as an unnecessary technical error and issue incorrect corrective instructions, thereby causing a serious safety accident.

[0086] This embodiment proposes an online method for detecting this coordinate system mismatch, the core of which lies in capturing the kinematic discrepancies between different sensors. The vision system measures the absolute motion of the foot relative to a fixed ground surface. The pressure-sensing pad, on the other hand, measures the relative motion of the pressure center relative to the pad itself. Under normal circumstances, these two motions should be highly correlated. However, when the pressure-sensing pad slips, the exerciser's foot slides along with the pad, and the vision system detects a significant horizontal velocity of the foot; but from the perspective of the pressure-sensing pad, there is no relative movement between the foot and the pad, therefore its reported pressure center velocity is close to zero.

[0087] By utilizing this contradiction, the absolute horizontal velocity vector of the key foot points and the relative horizontal velocity vector of the pressure center are first calculated separately. Then, the difference between these two velocity vectors, i.e., the velocity divergence, is calculated. This divergence increases significantly when the pressure sensing pad slips.

[0088] However, a large divergence alone is insufficient to determine slippage, as even rapid, voluntary foot movements can cause temporary speed differences. To differentiate between these two scenarios, trunk acceleration signals are introduced. If the rapid foot movement is actively initiated by the practitioner, their trunk will inevitably exhibit corresponding acceleration or deceleration; therefore, a strong correlation should exist between foot speed and trunk acceleration. Conversely, if the foot passively slips along with the pressure-sensing pad, this correlation will be weak.

[0089] Therefore, the final judgment rule is: when the velocity divergence exceeds a preset divergence threshold, and its correlation coefficient with the trunk acceleration signal is lower than a preset active motion threshold, a passive slip event is determined to have occurred. Once slip is determined to have occurred, the most critical response is to immediately block the output of the immediate correction command. This is because, in the case of coordinate system mismatch, all stability judgments based on pressure data are unreliable. Forcibly outputting correction commands at this time is highly likely to produce negative interference. By blocking the command and completely returning control to the practitioner's own physiological reflexes, it is the safest and most reasonable choice at this moment.

[0090] In a preferred embodiment, the method further includes the following steps: Monitor real-time motion data and support status data to identify dynamic time windows where all speed indicators are below the static threshold; Within a dynamic time window, the coordinates of multiple key points on the foot and the coordinates of the pressure center are collected as paired samples. Solve the two-dimensional rigid transformation parameters, including translation and rotation, based on paired samples; The coordinate transformation matrix, updated based on two-dimensional rigid transformation parameters, is used to map real-time motion data and support state data. This matrix is ​​used to transform the support state data into a coordinate system consistent with the real-time motion data during the process of determining the tilting moment of the user's body based on the real-time motion data and support state data. This process determines the boundary of the support area and calculates the lever arm distance of the body's center of mass projection point relative to the boundary of the support area, thereby calculating the tilting moment.

[0091] After detecting a slip event and executing a temporary command to block it, the system needs to restore the correct alignment of the coordinate system; otherwise, subsequent guidance will remain unreliable.

[0092] The core idea of ​​this embodiment is to perform self-calibration by utilizing the low-dynamic, relatively static moments during training breaks or movement transitions. First, by monitoring visual foot velocity and pressure center velocity, dynamic time windows are identified where both are below a certain static threshold. Within these time windows, the pressure sensing pad and the trainee's foot can be considered to be in a relatively stable state. At this time, the visually captured foot position and the pressure center position reported by the pressure sensing pad should have a fixed geometric correspondence.

[0093] Within these identified time windows, multiple sets of paired samples are collected; each set contains the coordinates of a visual foot keypoint and a corresponding pressure center coordinate. Once a sufficient number of paired samples are collected, the coordinate transformation relationship between the two can be solved. Assuming the slippage of the pressure sensing pad mainly involves translation and minute rotation, this transformation relationship can be described by a two-dimensional rigid transformation. Using optimization algorithms such as the least squares method, the translational and rotational amounts involved in this transformation can be calculated based on the collected paired samples.

[0094] After solving for the transformation parameters that describe the current coordinate system misalignment, the final step is to use them to update the coordinate transformation matrix stored in the system. This matrix is ​​a key parameter used to transform the local coordinates of the pressure sensing pad to the global visual coordinate system. By superimposing the newly solved translation and rotation amounts onto the original matrix, the online correction of the coordinate system is completed.

[0095] After the correction is completed, in subsequent tipping moment calculations, all data from the pressure sensor pad will first be transformed using this updated coordinate transformation matrix before being compared and fused with data from the vision system. This eliminates the coordinate system mismatch problem caused by pressure sensor pad slippage, allowing the entire guidance system to return to an accurate and reliable working state, ensuring the correctness of the tipping moment calculation, and thus providing a correct premise for subsequent compensation judgment and feedback control.

[0096] Secondly, referring to Figure 2 This application also proposes a real-time feedback and interactive guidance system for martial arts, including: The acquisition module 210 is used to acquire real-time motion data and support status data of the user during exercise; The first determining module 220 determines the tilting moment of the user's body based on real-time motion data and the support state data. The tilting moment is used to characterize the instability trend of the user's body relative to the support area. The second determining module 230 determines the correction torque generated by a specific limb segment of the user based on real-time motion data. The correction torque is used to characterize the contribution of the motion of the specific limb segment of the user to the recovery of the body's posture. The judgment module 240 determines whether the movement of a specific limb segment is a physiological compensatory movement or a technical error movement based on the directional and amplitude relationship between the correction torque and the tilting torque. Control module 250 controls the output strategy of the feedback output device based on the judgment result: When a physiological compensatory action is identified, a feedback inhibition strategy targeting a specific limb segment is executed to block the output of immediate corrective instructions. When a technical error is detected, a feedback strategy targeting specific limb segments is executed to output immediate corrective instructions.

[0097] By acquiring real-time motion and support status data of users during exercise, the system determines the user's tilting torque and the corrective torque generated by specific limb segments based on this data. Based on the direction and amplitude relationship between these two data points, it intelligently determines whether the movement of a specific limb segment is a physiological compensatory movement or a technical error. When it is determined to be a physiological compensatory movement, a feedback inhibition strategy is implemented to avoid interfering with the user's instinctive balance recovery; when it is determined to be a technical error, a feedback prompt strategy is implemented, outputting immediate corrective instructions. This effectively distinguishes between physiological compensatory movements and technical errors, avoiding misjudgment and negative intervention of physiological compensation, thereby improving the safety, effectiveness, and user experience of training.

[0098] Furthermore, in some preferred embodiments, the martial arts real-time feedback interactive guidance system proposed in this application can perform any of the steps in the above methods.

[0099] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A real-time feedback and interactive guidance method for martial arts, characterized in that, include: Acquire real-time motion data and support status data of the user during exercise; Based on the real-time motion data and the support status data, the tilting moment of the user's body is determined, and the tilting moment is used to characterize the instability trend of the user's body relative to the support area. Based on the real-time motion data, a correction torque generated by a specific limb segment of the user is determined. The correction torque is used to characterize the contribution of the motion of the specific limb segment of the user to the recovery of body posture. Based on the directional and amplitude relationship between the corrective torque and the tilting torque, it is determined whether the movement of the specific limb segment is a physiological compensatory movement or a technical error movement. The output strategy of the control feedback output device is determined based on the judgment result. When the physiological compensatory action is determined, a feedback inhibition strategy targeting the specific limb segment is executed to block the output of immediate corrective instructions; When the action is determined to be a technical error, a feedback prompt strategy targeting the specific limb segment is executed to output the immediate correction instruction.

2. The real-time feedback interactive guidance method for martial arts according to claim 1, characterized in that, The step of determining whether the movement of a specific limb segment is a physiological compensatory movement or a technical error movement based on the directional and amplitude relationship between the corrective torque and the tilting torque includes: The direction of the corrective torque is compared with the direction of the tilting torque, and the amplitude of the corrective torque is calculated as a first difference between the amplitude of the corrective torque and the preset effective compensation threshold. When the direction of the corrective torque is opposite to the direction of the tilting torque, and the first difference is greater than zero, the movement of the specific limb segment is determined to be the physiological compensatory movement. When the direction of the corrective torque is the same as the direction of the tilting torque, or when the first difference is less than or equal to zero, the movement of the specific limb segment is determined to be a technical error.

3. The real-time feedback interactive guidance method for martial arts according to claim 1, characterized in that, The step of determining the tilting torque of the user's body based on the real-time motion data and the support status data includes: The position of the user's total center of mass horizontally projected onto the ground plane is determined based on the real-time motion data. The ground support area boundary of the user's body is determined based on the support status data; Calculate the shortest horizontal distance from the horizontal projection point to the boundary of the ground support area, and determine the deviation direction corresponding to the shortest horizontal distance; The tipping moment is calculated based on the user's total body mass and gravitational acceleration.

4. The real-time feedback interactive guidance method for martial arts according to claim 1, characterized in that, The method also includes the following steps: The internal clock of the inertial measurement unit that collects the real-time motion data is set as the global time reference; For the data streams of the optical vision device that collects the real-time motion data and the pressure sensing device that collects the support state data, time backtracking interpolation processing is performed according to the global time reference to align the data streams of the optical vision device and the pressure sensing device to the time axis sampling points of the inertial measurement unit. The aligned data stream is filtered, and the length of the filtering window is adjusted according to the noise variance of the current signal.

5. The real-time feedback interactive guidance method for martial arts according to claim 4, characterized in that, The step of filtering the aligned data stream, wherein the length of the filtering window is adjusted according to the noise variance of the current signal, includes: The noise estimation window is formed by selecting the N most recent sampled values ​​of the current signal within the sliding circular buffer. For the sampled values ​​within the noise estimation window, a noise variance estimate σ² is calculated. The noise variance estimate σ² is determined by the average of the squared deviations of the sampled values ​​relative to the mean within the estimation window, or by the variance of adjacent sampled difference sequences. The noise variance estimate σ² is input into a preset monotonic mapping function f(·) to obtain the target filtering window length L-target, such that L-target increases when σ² increases and decreases when σ² decreases, and L-target is restricted between the preset minimum window length L-min and maximum window length L-max. When |L-target−L-curr| is less than the preset change threshold ΔL, the current window length L-curr remains unchanged, or the window length is updated only when L-target satisfies the same increasing / decreasing trend within M consecutive noise estimation windows; When the update condition is met, the current filter window length is updated to L-new, and the filtering process is performed based on the updated window length.

6. The real-time feedback interactive guidance method for martial arts according to claim 1, characterized in that, The step of determining the tilting torque of the user's body based on the real-time motion data and the support status data includes: Real-time monitoring of the user's torso angular velocity; When the torso angular velocity is lower than a preset dynamic threshold, a preset standard weight allocation strategy is used to fuse the real-time motion data and the support state data to calculate the body's center of mass position. When the torso angular velocity exceeds the dynamic threshold, the trust weight of the visual data in the real-time motion data is reduced, and the trust weight of the inertial data and the support state data in the real-time motion data is increased. The body center of mass position is calculated based on the adjusted trust weight, and the tilting torque is calculated based on the body center of mass position.

7. The real-time feedback interactive guidance method for martial arts according to claim 6, characterized in that, The steps of calculating the body's center of mass position based on the adjusted trust weights and calculating the tipping moment based on the body's center of mass position include: Obtain the horizontal projection position of the body's center of mass P-com-vis calculated based on visual data, and the horizontal projection position of the body's center of mass P-com-imu calculated based on inertial data; Obtain the adjusted visual trust weight w-vis and inertial trust weight w-imu, and normalize the trust weights to satisfy: w-vis + w-imu = 1; The centroid horizontal projection positions are weighted and fused according to the normalized trust weights to obtain the fused centroid horizontal projection position P-com, wherein the weighted fusion satisfies: P-com = w-vis P-com-vis + w-imu P-com-imu; The boundary of the ground support area is determined based on the support status data, and the shortest horizontal distance d from the horizontal projection position P-com of the fused centroid to the boundary of the ground support area is calculated. The horizontal deviation direction of the horizontal projection position of the fused centroid relative to the boundary of the ground support area is then determined. The magnitude of the tipping moment, abs(M-tip), is calculated based on the user's total body mass m-total and gravitational acceleration g. The magnitude of the tipping moment satisfies the following: abs(M-tip) = m-total g d; And make the direction of the tilting moment consistent with the direction of the horizontal deviation.

8. The real-time feedback interactive guidance method for martial arts according to claim 1, characterized in that, The method also includes the following steps: Calculate the absolute horizontal velocity vector of the key points of the foot in the real-time motion data, and the relative horizontal velocity vector of the center of pressure in the support state data; Calculate the velocity divergence between the absolute horizontal velocity vector and the relative horizontal velocity vector; Calculate the correlation coefficient between the velocity divergence and the trunk acceleration signal in the real-time motion data; When the velocity divergence exceeds a preset divergence threshold and the correlation coefficient is lower than a preset active motion threshold, a passive slip event is determined to have occurred, and the output of the immediate correction command is blocked.

9. The real-time feedback interactive guidance method for martial arts according to claim 1, characterized in that, The method also includes the following steps: Monitor the real-time motion data and the support status data to identify dynamic time windows where all speed indicators are below the static threshold; Within the dynamic time window, multiple sets of coordinates of the key points of the foot and the coordinates of the pressure center are collected as paired samples. Based on the paired samples, solve for the two-dimensional rigid transformation parameters, which include translation and rotation. The coordinate transformation matrix used to map the real-time motion data and the support state data is updated based on the two-dimensional rigid transformation parameters. This is used to transform the support state data to a coordinate system consistent with the real-time motion data during the process of determining the tilting moment of the user's body based on the real-time motion data and the support state data. This allows for the determination of the support area boundary and the calculation of the lever arm distance of the body's center of mass projection point relative to the support area boundary, thereby calculating the tilting moment.

10. A real-time feedback interactive guidance system for martial arts, used to execute the solution according to any one of claims 1 to 9, characterized in that, include: The acquisition module is used to acquire real-time motion data and support status data of the user during exercise; The first determining module determines the tilting moment of the user's body based on the real-time motion data and the support state data. The tilting moment is used to characterize the instability trend of the user's body relative to the support area. The second determining module determines the correction torque generated by a specific limb segment of the user based on the real-time motion data. The correction torque is used to characterize the contribution of the motion of the specific limb segment of the user to the recovery of the body's posture. The determination module determines whether the movement of the specific limb segment is a physiological compensatory movement or a technical error movement based on the directional and amplitude relationship between the correction torque and the tilting torque. The control module controls the output strategy of the feedback output device based on the judgment result. When the physiological compensatory action is determined, a feedback inhibition strategy targeting the specific limb segment is executed to block the output of immediate corrective instructions; When the action is determined to be a technical error, a feedback prompt strategy targeting the specific limb segment is executed to output the immediate correction instruction.

Images