A cognitive judgment training device for multi-mode movement

Abstract

The present application relates to the technical field of intelligent wear and motion evaluation, and discloses a cognitive judgment training device for multi-mode motion, which comprises a central monitoring unit at the chest and a limb sensing unit, and the central monitoring unit is used for establishing a trunk reference system. A processing unit is configured to use a limb distribution structure parameter as a force arm, combine the relative motion acceleration of the limb relative to the trunk, and generate a limb-trunk correlation torque signal representing body feedforward posture control; according to the time sequence fluctuation of the signal, the main action triggering moment is locked, and the expected posture adjustment cumulative amount of the signal on the horizontal reference plane within a preset time window before the moment is calculated; further, the kinetic energy distribution proportion of the cumulative amount in the total amount of whole process dynamics including the action burst period is calculated, and an evaluation result is generated accordingly. The present application can effectively penetrate the interference of large amplitude pre-swing action, accurately quantify the weak feedforward posture adjustment before starting, and realize unified evaluation of the concentration degree of multi-mode motion.

Description

Technical Field

[0001] This invention relates to the field of smart wearable and motion assessment technology, and more specifically, to a cognitive judgment training device for multimodal motion. Background Technology

[0002] In competitive sports and rehabilitation medicine, high-quality execution of motor skills often depends on highly focused attention before the initiation of a movement. Research shows that high-level athletes engage in cognitive preparation through mechanisms such as eye stillness in the short time before initiating a movement. This cognitive preparation manifests in body dynamics as extremely subtle feedforward postural adjustments. These adjustments are a key element in ensuring the accuracy and stability of the movement.

[0003] Currently, digital assessment of motion processes primarily relies on inertial measurement units (IMUs) to capture motion trajectories and judge training quality by comparing the smoothness, consistency, or standardization of the trajectories. However, this existing technology based on full-segment trajectory generation has significant technical limitations when assessing attention concentration, a specific cognitive indicator:

[0004] First, attentional concentration often occurs within a very short time window before the initiation of an action, constituting a short-phase event. The corresponding body posture adjustments are minute and often accompanied by high-frequency discrete oscillations. Existing trajectory generation algorithms typically filter and smooth the collected data to obtain smooth motion curves. This processing method easily treats the minute but crucial posture adjustment signals during the preparation period as noise and filters them out, resulting in the dilution or smoothing of data segments containing cognitive preparation information, failing to effectively capture the true psychological and physiological readiness state before the initiation of the action.

[0005] Secondly, in multi-modal sports scenarios involving throwing, swinging, and jumping, the movements themselves naturally include large-amplitude pre-swing, power-building, or reverse loading actions. These mechanical actions, determined by the movement mode, highly overlap with the attention preparation period in time, and the acceleration and amplitude they generate are far greater than those of minute posture adjustment signals. Existing single-threshold or trajectory morphology analysis techniques struggle to extract weak cognitive preparation signals from large-amplitude power-building actions. This makes the system highly susceptible to misjudging large-amplitude mechanical power-building as high-quality attention preparation, resulting in false alarms and an inability to provide a consistent and accurate cognitive state determination across multiple modes.

[0006] In summary, existing motion assessment devices struggle to accurately extract cognitive readiness information within a short pre-start window from strong background noise during multimodal motion. Summary of the Invention

[0007] This invention provides a cognitive judgment training device for multimodal motion, which solves the technical problems mentioned in the background art.

[0008] The present invention provides a cognitive judgment training device for multimodal movement, the device comprising a central monitoring unit worn on the user's sternum, limb sensing units worn on the user's limbs, and a data processing unit.

[0009] The central monitoring unit is used to establish a human torso reference system with the chest as the reference point and to lock the gravity reference direction.

[0010] The data processing unit is configured to perform the following steps:

[0011] Based on the limb distribution structure parameters of the limb sensing unit in the human torso reference frame as the lever arm, and combined with the relative motion acceleration of the limb sensing unit relative to the central monitoring unit, a limb-torso related torque signal characterizing the body feedforward posture control is generated.

[0012] Based on the temporal fluctuation of the associated torque signal, the main action trigger time is locked, and within a preset preparation time window before the main action trigger time, the expected cumulative amount of posture adjustment of the associated torque signal on the horizontal posture reference plane perpendicular to the gravity reference direction is calculated.

[0013] The proportion of the expected cumulative posture adjustment in the total dynamics of the entire process, including the burst phase of the movement, is calculated. Based on the degree of matching between the proportion of kinetic energy distribution and the preset target interval, the attention concentration evaluation results of multi-mode movement are generated.

[0014] The beneficial effects of this invention include: by utilizing an array layout of a central unit worn on the sternum and sensing units at the ends of the limbs, and leveraging the mechanical amplification effect of the long lever arm of the limb on the micro-motion data at the ends, it effectively captures weak posture adjustment signals within a very short time window before the start of the movement, solving the problem that conventional trajectory algorithms easily smooth out key cognitive information; at the same time, by calculating the proportion of posture adjustment during the preparation period in the total dynamics of the entire process, it distinguishes between cognitively driven posture preparation and the mechanical power accumulation inherent in the movement pattern, thereby achieving anti-interference, high sensitivity, and unified evaluation of the degree of attention concentration in multi-mode movement. Attached Figure Description

[0015] Figure 1 This is a flowchart of the process of a multi-modal motion cognitive judgment training device according to the present invention;

[0016] Figure 2 This is a schematic diagram of a wearable cognitive judgment training device for multimodal motion according to the present invention. Detailed Implementation

[0017] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0018] like Figure 1 As shown, a multimodal motion cognitive judgment training device includes a central monitoring unit worn on the user's sternum, limb sensing units worn on the user's limbs, and a data processing unit.

[0019] The central monitoring unit is used to establish a human torso reference system with the chest as the reference point and to lock the gravity reference direction.

[0020] The data processing unit is configured to perform the following steps:

[0021] Based on the limb distribution structure parameters of the limb sensing unit in the human torso reference frame as the lever arm, and combined with the relative motion acceleration of the limb sensing unit relative to the central monitoring unit, a limb-torso related torque signal characterizing the body feedforward posture control is generated.

[0022] Based on the temporal fluctuation of the associated torque signal, the main action trigger time is locked, and within a preset preparation time window before the main action trigger time, the expected cumulative amount of posture adjustment of the associated torque signal on the horizontal posture reference plane perpendicular to the gravity reference direction is calculated.

[0023] The proportion of the expected cumulative posture adjustment in the total dynamics of the entire process, including the burst phase of the movement, is calculated. Based on the degree of matching between the proportion of kinetic energy distribution and the preset target interval, the attention concentration evaluation results of multi-mode movement are generated.

[0024] like Figure 2 As shown, preferably, the central monitoring unit worn on the user's sternum, the limb sensing units worn on the user's extremities, and the data processing unit include:

[0025] The central monitoring unit is worn above the xiphoid process of the user's sternum, serving as the origin of the human torso reference system. ;

[0026] The limb sensing unit They are worn on the back of the user's left wrist joint, the back of the right wrist joint, the outside of the left ankle joint, and the outside of the right ankle joint, respectively.

[0027] The central monitoring unit is worn on the specific part of the user's sternum. Preferably, it is placed 2 to 3 centimeters above the xiphoid process of the sternum. Using the xiphoid process as a clear anatomical landmark, this position reduces interference from trunk movements on the reference frame origin, ensuring consistent positioning each time it is worn.

[0028] The number of limb sensing units is the total number of limb-end devices participating in data acquisition. Preferably, there are 4 units, corresponding to the four key movement parts of the human body's limbs, which can completely construct a distributed array centered on the chest, covering the limbs related to major posture adjustments throughout the body.

[0029] The limb sensing units are worn at fixed locations on the body. Preferably, they are located 5 mm from the ulnar styloid process on the dorsal side of the left wrist joint, 5 mm from the ulnar styloid process on the dorsal side of the right wrist joint, 8 mm from the lateral malleolus on the lateral side of the left ankle joint, and 8 mm from the lateral malleolus on the lateral side of the right ankle joint. These locations are close to bony landmarks, making it less likely for the sensors to shift during movement and maximizing the lever arm length of the limb relative to the chest, which is beneficial for capturing weak posture signals.

[0030] The origin of the human torso reference frame is the benchmark point for constructing the human torso reference frame. Specifically, it is the point on the skin surface where the geometric center of the central monitoring unit's outer shell is projected vertically. This point coincides with the wearing position of the central monitoring unit, providing a unified benchmark for all subsequent spatial parameter calculations.

[0031] The limb sensing unit number is a unique identifier for the four limb sensing units. Preferably, it is the right arm (RA), left arm (LA), right leg (RL), and left leg (LL).

[0032] The central monitoring unit is positioned 2-3 cm above the xiphoid process of the sternum. Four limb sensing units are fixed to the dorsal side of the left and right wrist joints and the lateral side of the left and right ankle joints, forming a distributed array centered on the chest. Conventional wearable devices may only place sensors on a single part of the limb or torso, but this layout specifically considers the need to capture posture adjustment signals. For example, the wrist and ankle joints, as limb extremities, have the longest lever arms relative to the chest, amplifying minute limb accelerations into significant torque signals, facilitating the subsequent detection of subtle posture adjustments before the start of an action. Simultaneously, this layout ensures that the four limb sensing units collect data in a unified reference frame, providing a structural basis for subsequent calculation of the associated torque signals between the limbs and torso, avoiding reference frame confusion caused by dispersed sensor positions.

[0033] The central monitoring unit is secured using an elastic bandage. The bandage should be tight enough to ensure a seamless fit between the sensor and skin without obstructing breathing or trunk movement. After securing, the sensor displacement should not exceed 2 mm to ensure stable origin positioning. The limb sensing unit uses an adhesive mounting base combined with medical tape for double fixation. There should be no air gap between the mounting base and the skin, and the sensor displacement relative to the limb should not exceed 1 mm during movement. For example, when fixing the sensor on the dorsal side of the left wrist joint, first clean off any skin oils, then attach the mounting base 5 mm from the ulnar styloid process, and finally wrap two turns of medical tape around the edge of the mounting base for reinforcement, preventing it from falling off or shifting during movement. When wearing the unit, ensure that the geometric center of the central monitoring unit completely coincides with the projection point 2-3 cm above the xiphoid process of the sternum, and that the center of the limb sensor deviates from the preset wearing position by no more than 3 mm to ensure consistent positioning of all sensors.

[0034] Preferably, a human torso reference frame is established with the chest as the reference point, and the gravity reference direction is locked, including:

[0035] The user is instructed to remain stationary while standing for an initialization period, during which the central monitoring unit collects triaxial acceleration data. And calculate its arithmetic mean. ;

[0036] The arithmetic mean The normalized vector is defined as the gravity reference direction. The calculation formula is:

[0037]

[0038] Will be in relation to the gravity reference direction The opposite direction is defined as the vertical axis direction of the human torso reference frame. The calculation formula is:

[0039]

[0040] The forward axis direction of the human torso reference system is determined based on the preset wearing orientation of the central monitoring unit. ;

[0041] via the forward axis direction With the vertical axis direction The vector cross product operation is used to determine the transverse axis direction of the human torso reference frame. The calculation formula is:

[0042]

[0043] Thus, an orthogonal human torso reference system is constructed. .

[0044] The initialization period is a specific time interval during which the user remains stationary to establish a reference frame. A period of 2 seconds is preferred, as this duration balances data stability with user ease of operation, and is sufficient to collect stable acceleration data for accurate estimation of the direction of gravity.

[0045] The triaxial acceleration data from the central monitoring unit refers to the acceleration data in the x, y, and z axes collected by the central monitoring unit during the initialization period and during motion. This data can be acquired through the triaxial microelectromechanical system (MEMS) accelerometer sensor built into the central monitoring unit.

[0046] The triaxial acceleration arithmetic mean is the arithmetic average of all triaxial acceleration data within the initialization period. It is used for subsequent calculations in the gravity reference direction, reflecting the average level of acceleration during the initialization period.

[0047] The gravity reference direction is the vector direction obtained by normalizing the arithmetic mean of the three-axis accelerations. This direction coincides with the actual gravity direction and provides a reference for constructing a frame of reference.

[0048] The vertical axis of the human torso reference frame is opposite to the direction of the gravitational reference. This vertical direction serves as the reference frame's reference point, used to distinguish between up and down.

[0049] The forward axis of the human torso reference frame is determined based on the preset wearing orientation of the central monitoring unit. Preferably, the direction pointed to by the arrow on the outer shell of the central monitoring unit is consistent with the front of the human body, in order to conform to the conventional orientation definition for human motion posture assessment and ensure directional consistency.

[0050] The transverse axis of the human torso reference frame is obtained by performing a vector cross product between the forward axis and the vertical axis. This direction, together with the forward and vertical axes, forms an orthogonal coordinate system used to distinguish left and right directions.

[0051] The human torso reference system is an orthogonal three-dimensional coordinate system formed by taking the central monitoring unit's wearing position as the origin and combining it with the vertical axis, forward axis, and lateral axis. It provides a unified reference framework for the calculation of all subsequent spatial parameters and motion data.

[0052] This method collects triaxial acceleration data from the central monitoring unit during a static standing initialization period, calculates its arithmetic mean, and normalizes it to obtain the gravity reference direction. Then, combined with the preset wearing orientation, the forward axis is determined, and the lateral axis is obtained through vector cross product, ultimately constructing an orthogonal human torso reference frame. Conventional human reference frame construction may rely on external calibration equipment or complex algorithms, while this solution can complete the process using only data collected by the central monitoring unit itself. For example, before a user practices shooting, they remain static for 2 seconds. The central monitoring unit collects triaxial acceleration data during this time, calculates the average, normalizes it to obtain the gravity reference direction, and then determines the forward axis as the front of the body based on the device's arrow direction. The lateral axis, obtained through cross product, represents the left and right directions of the body, thus constructing a stable reference frame.

[0053] The default wearing orientation of the central monitoring unit is that the arrow marked on the device casing is perfectly aligned with the front of the human body, with an allowable alignment error of no more than 5 degrees, ensuring consistency in the forward axis direction. The system sampling frequency must be at least 100 Hz. This sampling frequency meets the accuracy requirements for gravity direction estimation and avoids data distortion due to excessively low sampling frequencies. The triaxial acceleration data collected during the initialization period must undergo low-pass filtering with a cutoff frequency of 5 Hz to remove high-frequency noise interference from the arithmetic mean calculation. For example, a finite impulse response filter can be used to process the raw acceleration data, filtering out motion interference and environmental noise to ensure that the calculated triaxial acceleration arithmetic mean accurately reflects the gravity direction.

[0054] Preferably, based on the limb distribution structure parameters of the limb sensing unit in the human torso reference frame as the lever arm, and combined with the relative motion acceleration of the limb sensing unit relative to the central monitoring unit, a limb-torso associated torque signal characterizing body feedforward posture control is generated, including:

[0055] The spatial position vector of the limb sensing unit relative to the central monitoring unit is obtained in the user's standard upright initial posture, and is defined as the initial lever arm vector in the human torso reference frame. ;

[0056] Get user weight data And using a preset limb segment mass ratio coefficient Calculate the corresponding limb sensing unit Limb segment quality The calculation formula is:

[0057]

[0058] The real-time acceleration collected by the limb sensing unit via attitude rotation matrix Mapped to the human torso reference frame, and the acceleration of the central monitoring unit in the human torso reference frame is subtracted. The relative acceleration is obtained. The calculation formula is:

[0059]

[0060] For each of the limb sensing units, the initial lever arm vector is calculated. With the relative acceleration The cross product of the vectors, multiplied by the corresponding mass of the limb segment. Sum all results to generate the limb-trunk associated torque signal. The calculation formula is:

[0061]

[0062] in This represents the vector cross product operation.

[0063] The standard upright initialization posture is the standard static posture that the user maintains to obtain the initial lever arm vector. The preferred posture is to stand with feet shoulder-width apart, arms hanging naturally at the sides with palms against the outside of the thighs, and head upright. This posture ensures the stability of the limbs relative to the torso, facilitating accurate acquisition of the spatial position vector.

[0064] The initial lever arm vector is the three-dimensional spatial position vector of the limb sensing unit relative to the central monitoring unit under a standard upright initialization posture. It can be calculated by measuring the spatial distance and direction between the limb sensing unit and the central monitoring unit using a laser rangefinder and combining this with the human torso reference frame.

[0065] User weight data is fundamental data reflecting a user's overall health. It can be obtained either manually entered by the user through the device interface or automatically when the device is connected to a scale.

[0066] The limb segment mass ratio coefficient is a preset fixed proportion of the mass of each limb segment to the user's body weight. The preferred ratios are 0.028 for the right arm, 0.028 for the left arm, 0.103 for the right leg, and 0.103 for the left leg. This ratio conforms to the conventional range of limb segment mass to body weight in adult human anatomy and can approximately reflect the actual mass of the limb segments.

[0067] Limb segment mass is an estimate of the mass of a single limb segment. It is obtained by multiplying the user's weight data by the corresponding limb segment mass ratio coefficient, and is used to quantify the mechanical contribution of the limb segment.

[0068] The real-time acceleration of the limb sensing unit refers to the three-axis acceleration data in its own coordinate system collected by the limb sensing unit during movement. This data can be acquired in real time using the three-axis microelectromechanical system (MEMS) accelerometer sensor built into the limb sensing unit.

[0069] The attitude rotation matrix is ​​used to transform the acceleration data of the limb sensing unit's own coordinate system to the human torso reference system. It is obtained by calibrating the attitude of the limb and the central monitoring unit during the initialization period to ensure the consistency of data in different coordinate systems.

[0070] The acceleration of the central monitoring unit in the torso reference frame is the result of converting the acceleration data collected by the central monitoring unit to the human torso reference frame. It is used to calculate the motion acceleration of the limbs relative to the torso.

[0071] Relative acceleration is the acceleration data of a limb relative to the central monitoring unit. It eliminates interference from overall translational motion and focuses on the independent movement of the limb relative to the torso.

[0072] The single-limb torque component is the torque contribution of a single limb to the trunk. It is calculated from the initial lever arm vector, relative acceleration, and limb segment mass.

[0073] The limb-trunk associated torque signal is an overall mechanical signal obtained by integrating the torque components of all limbs. It represents the overall state of body feedforward posture control.

[0074] This method uses the limb spatial position vector in a standard upright initial posture as the initial lever arm vector. Combined with relative motion acceleration and limb segment mass, a complete process is employed to generate a limb-torso correlated torque signal through vector cross product summation. Conventional motion assessments often directly use raw trajectory or acceleration data. However, this approach, through the coupled calculation of lever arm, acceleration, and mass, amplifies the minute accelerations at the limb extremities into detectable torque signals. For example, during the shooting preparation phase, the minute adjustment acceleration of the wrist joint is transformed into a significant torque change through the long lever arm of the arm, avoiding being drowned out by environmental noise. Simultaneously, the calculation method for relative motion acceleration (first rotating and mapping to the torso reference frame, then subtracting the acceleration from the central monitoring unit) specifically eliminates interference from overall body translation, ensuring that the data only reflects the limb's posture adjustment relative to the torso.

[0075] The initial lever arm vector measurement must specify that the sampling frequency of the laser rangefinder sensor is no less than 100 Hz, and the measurement accuracy error is no more than 2 mm to ensure the accuracy of the spatial position vector. The limb segment mass ratio coefficient must be specified as applicable to adults aged 18 to 60. For children or users with special body types, adjustments can be made based on age, height, and weight; the correction coefficient can be retrieved from the built-in human proportion database. The calibration process for the posture rotation matrix must be detailed: During the initialization period, the user maintains a standard upright posture. The device simultaneously collects posture data from the central and limb sensing units, calculates the rotation matrix using a quaternion conversion algorithm, and the limbs must remain stationary during calibration to avoid motion interference. The time synchronization error between the central monitoring unit and the limb sensing unit must be controlled within 1 millisecond, which can be achieved through internal clock calibration to ensure that acceleration data match at the same point in time, avoiding calculation errors.

[0076] Preferably, locking the main action trigger time based on the timing fluctuation of the associated torque signal includes:

[0077] Construct a direction perpendicular to the gravity reference direction Horizontal attitude reference plane projection matrix and the limb-torso associated torque signal Projecting this onto the plane yields the horizontal projection moment vector. The calculation formula is:

[0078]

[0079]

[0080] Calculate the magnitude of the horizontal projected moment vector to obtain the real-time start-up intensity signal. The calculation formula is:

[0081]

[0082] Using the initialization period The arithmetic mean of the background noise is calculated from the real-time start-up intensity signal. and standard deviation ;

[0083] Based on the arithmetic mean The standard deviation and preset threshold coefficients Build a startup threshold The calculation formula is:

[0084]

[0085] Scan the real-time startup intensity signal in chronological order. The first consecutive threshold exceeding the start-up determination threshold will be... And the number of continuous samples reaches the preset number of samples. The moment is locked as the trigger moment of the main action. The calculation formula is:

[0086]

[0087] The horizontal posture reference plane projection matrix is ​​used to project the limb-torso associated torque signal onto a plane perpendicular to the gravity reference direction. It is obtained by subtracting the product of the gravity reference direction vector and its transpose from a 3rd-order identity matrix, thus focusing the signal on the horizontal plane.

[0088] The horizontal projected moment vector is obtained by processing the limb-torso related moment signal through the projection matrix of the horizontal posture reference plane. Only the moment components related to posture stability in the horizontal direction are retained, while vertical interference is eliminated.

[0089] The real-time start-up intensity signal is the magnitude of the horizontally projected torque vector. It is used to quantify the intensity changes of the torque signal, reflecting the mechanical fluctuations before the action begins.

[0090] The background noise arithmetic mean is the arithmetic average of the real-time start-up intensity signals during the initialization period. It characterizes the noise level at initialization and provides a reference for threshold setting.

[0091] The background noise standard deviation is the standard deviation of the real-time start-up intensity signal during the initialization period. It reflects the dispersion of the noise and helps to construct a reasonable start-up judgment threshold.

[0092] The threshold coefficient is a preset constant used to adjust the sensitivity of the start-up judgment threshold. It is preferably 2.5, as this value can balance the false trigger rate and the missed trigger rate, and adapt to the noise level of most motion scenarios.

[0093] The start-up threshold is a critical signal strength value used to identify active start-up. It is obtained by adding the arithmetic mean of the background noise to the product of the threshold coefficient and the standard deviation of the background noise, thus distinguishing noise from valid start-up signals.

[0094] The continuous sample count is the number of samples for which the real-time start intensity signal must continuously exceed the start-up judgment threshold. Ten samples are preferred, to ensure, in conjunction with the system sampling frequency, that the trigger moment corresponds to the actual action start, avoiding false triggering due to single noise.

[0095] The trigger moment for the active action is the moment when the real-time start-up strength signal continuously exceeds the start-up judgment threshold and reaches the required number of continuous samples. This signifies that the active action is about to erupt and provides a time reference for defining the subsequent preparation window.

[0096] A projection matrix perpendicular to the gravity reference direction is constructed to project the associated torque signal onto a horizontal plane. A threshold is then established through real-time noise statistics of the start-up intensity signal. The moment when the main action is locked is triggered by consecutive samples exceeding the threshold. Conventional methods often directly use peak limb acceleration or a single threshold for triggering, which is easily affected by noise or pre-swinging motions. For example, in throwing motions, local acceleration peaks may occur during the pre-swinging phase, which conventional methods easily misjudge as the start-up moment. This scheme, however, focuses on the posture-stabilizing torque through horizontal projection, dynamically sets the threshold based on noise statistics during the initialization period, and requires 10 consecutive samples to exceed the threshold, effectively eliminating pre-swinging interference and instantaneous noise. This design specifically addresses the complexity of the start-up signal in multi-mode motion, ensuring that the trigger moment accurately corresponds to the node where the overall posture chain enters the execution state, providing an accurate time reference for calculating the cumulative amount of expected posture adjustment within the subsequent short window.

[0097] The calculation of the horizontal posture reference plane projection matrix must clearly define the 3rd order identity matrix as a matrix with diagonal elements of 1 and all other elements of 0, ensuring unambiguous projection logic. When calculating the arithmetic mean and standard deviation of background noise during the initialization period, the real-time startup intensity signal must first be low-pass filtered. The preferred filter cutoff frequency is 8 Hz, implemented using an infinite impulse response filter to remove the influence of high-frequency noise on the statistical results. The value of the continuous sample number must match the system sampling frequency. If the sampling frequency is 100 Hz, 10 samples correspond to 0.1 seconds. This duration avoids accidental triggering by instantaneous noise without delaying the startup locking. The calculation of the startup judgment threshold must clearly define the calculation range of the arithmetic mean and standard deviation as the entire initialization period, ensuring that the threshold reflects the noise level of the current environment. For example, if the initialization period is 2 seconds and the sampling frequency is 100 Hz, the mean and standard deviation are calculated based on 200 real-time startup intensity signal samples, and then combined with a threshold coefficient of 2.5 to obtain the startup judgment threshold.

[0098] Preferably, within a preset preparation time window prior to the main action triggering moment, the expected cumulative posture adjustment of the associated torque signal on a horizontal posture reference plane perpendicular to the gravity reference direction is calculated, including:

[0099] Get the pre-set preparation window duration parameter And combined with the system sampling frequency Determine the number of discrete sampling points corresponding to the preparation window duration parameter. The calculation formula is:

[0100]

[0101] Based on the main action trigger time Starting from the beginning, backtracking in the negative direction of the time axis, select multiple discrete moments covering the preparation window duration parameter, and calculate the expected cumulative posture adjustment. The calculation formula is:

[0102]

[0103] in:

[0104] For at any time The magnitude of the horizontal projected moment vector;

[0105] The time interval for a single sampling.

[0106] The preparation window duration parameter is a pre-set short time interval parameter used to capture posture adjustment before the start of an action. It is preferably 50 milliseconds to 200 milliseconds, as this range matches the short-phase characteristics of feedforward posture adjustment before the start of an action, which can fully cover the posture preparation process related to attention concentration, while avoiding the inclusion of irrelevant time period data.

[0107] The system sampling frequency is the number of times the device samples data such as acceleration and torque per unit time. It is preferably no less than 100 Hz. This frequency can ensure that there are enough discrete sampling points within a short preparation window, accurately reconstruct the temporal changes of the torque signal, and avoid signal distortion caused by sparse sampling.

[0108] The number of discrete sampling points in the preparation window is an integer number of sampling points calculated from the preparation window duration parameter and the system sampling frequency. It is used to specify the amount of discrete-time data to be included in the cumulative calculation, ensuring that the calculation range accurately corresponds to the preparation window.

[0109] The expected cumulative posture adjustment is the discrete integral result of the magnitude of the horizontal projected moment vector within the preparation window. Quantifying the total effect of body posture adjustment in the short phase before the start of the action is a parameter characterizing attention-related feedforward posture adjustments.

[0110] Starting from the moment the main action is triggered, the timeline is traced back in the negative direction. The number of sampling points is determined by combining the preset preparation window length and the system sampling frequency. The magnitude of the horizontal projected moment vector within this interval is accumulated and converted into a cumulative value. Conventional methods often use a fixed starting time window or integrate all data, which can easily include irrelevant data after the action begins or omit key preparation period signals. For example, in shooting, the moment the main action is triggered is the instant the trigger is pulled. Tracing back 100 milliseconds to a preparation window precisely covers the final posture calibration stage after aiming. Accumulating the magnitude of the horizontal projected moment vector during this period can accurately quantify posture stability adjustments related to attention concentration. This design, where the time window setting is strongly tied to the trigger moment, is specifically designed for the short-phase characteristics of attention, ensuring that the cumulative value only reflects the key posture preparation before the action begins, rather than the average effect of the entire movement trajectory.

[0111] The value of the preparation window duration parameter needs to be adjusted according to the sports mode. For fast-paced, explosive sports such as throwing, a value of 50 to 100 milliseconds is preferred, while for precision-controlled sports such as golf putting, a value of 150 to 200 milliseconds is preferred, adapting to the time characteristics of different sports preparation periods. The system sampling frequency must be clearly defined with a minimum of 100 Hz and a maximum of 200 Hz. Higher frequencies can improve signal resolution, but the power consumption and data storage pressure of the device must be balanced. When backtracking the preparation window, if the amount of effective data is insufficient for the number of discrete sampling points, the starting time of the device's data acquisition is taken as the starting point of the preparation window, and the window duration is not shortened to ensure the integrity of the cumulative calculation. Before accumulating the magnitude of the horizontal projected moment vector, the magnitude data at each discrete moment needs to be low-pass filtered. The filter cutoff frequency is 5 Hz, implemented using a finite impulse response filter to remove high-frequency noise interference to the accumulation result.

[0112] Preferably, calculating the proportion of the expected cumulative posture adjustment in the total kinetic energy distribution of the entire process, including the burst phase of the movement, includes:

[0113] Get the pre-set execution window duration parameter And in combination with the system sampling frequency Determine the number of execution sampling points corresponding to the execution window duration parameter. The calculation formula is:

[0114]

[0115] Based on the main action trigger time Starting from the positive direction of the time axis, calculate the dynamics of the action burst period within the execution window duration parameter. The calculation formula is:

[0116]

[0117] Accumulated amount of the expected posture adjustment With the dynamics of the burst phase of the action The total dynamics of the entire process are summed to obtain the cumulative amount of the expected posture adjustment. The proportion of kinetic energy distribution is obtained by taking the ratio of the total dynamics in the entire process. The calculation formula is:

[0118]

[0119] The execution window duration parameter is a pre-set time interval parameter used to capture the dynamic characteristics of the burst phase of motion. It is preferably 300 milliseconds to 1000 milliseconds, as this range can fully cover the burst phase of most multi-mode motions, without missing key dynamic information or including irrelevant decay motion data after the burst.

[0120] The number of execution sampling points is an integer number of sampling points calculated from the execution window duration parameter and the system sampling frequency. It is used to clarify the amount of discrete-time data that needs to be included in the calculation of dynamic quantities during the burst phase of the action, ensuring that the calculation range accurately corresponds to the burst phase.

[0121] The dynamic quantities during the burst phase of an action are the discrete integral results of the magnitude of the horizontally projected moment vector within the execution window. Quantifying the total mechanical quantities during the burst phase of an action, together with the accumulated quantities during the preparation phase, constitutes the dynamic basis of the entire process.

[0122] The total dynamics of the entire process is the sum of the expected cumulative amount of posture adjustment and the dynamics during the explosive phase of the movement. It reflects the total mechanical effect of the preparation and explosive phases of the movement and provides the denominator for the percentage calculation.

[0123] The kinetic energy distribution ratio is the ratio of the expected cumulative amount of posture adjustment to the total dynamics throughout the entire process. It characterizes the proportion of the mechanical contribution of posture adjustment during the action preparation phase to the total mechanical effect and is a core indicator for removing pre-swing disturbances.

[0124] Starting from the moment the main action is triggered, the sampling method proceeds in the positive direction of the time axis. The number of sampling points is determined by combining the preset execution window duration and the system sampling frequency. The magnitude of the horizontal projected moment vector within this interval is accumulated and converted into the dynamic quantity of the burst phase. This is then summed with the accumulated quantity during the preparation phase to calculate the proportion. Conventional methods often only focus on the preparation phase data or the total dynamic quantity of the entire phase, making it difficult to distinguish between cognitively driven posture preparation and the inherent mechanical energy storage of the pattern. For example, in a racket swing, the mechanical signals of the pre-swing and the burst of the swing are superimposed. Conventional methods cannot isolate the interference of the pre-swing. However, this solution, by calculating the proportion of the accumulated quantity during the preparation phase in the total mechanical quantity, can highlight the mechanical contribution corresponding to cognitive preparation and weaken the influence of inherent movements of the pattern, such as the pre-swing. This targeted proportion design specifically addresses the evaluation ambiguity problem of multi-mode sports.

[0125] The execution window duration parameter needs to be adjusted according to the sports mode. For fast-paced, explosive sports like basketball shooting, a duration of 300 to 500 milliseconds is preferred, while for sports with longer motions like a golf swing, a duration of 800 to 1000 milliseconds is preferred, adapting to the time characteristics of different sports' explosive phases. The system sampling frequency must be consistent with that used during the preparation phase calculation, preferably 100 to 200 Hz, to ensure uniform data resolution. If the execution window extends to the actual end time of the action before reaching the preset duration, the end time of the action is taken as the end point of the execution window, and the window is not extended further to avoid including irrelevant data. Before accumulating the magnitude of the horizontal projected moment vector, the same low-pass filtering process as during the preparation phase must be used, with a filter cutoff frequency of 5 Hz, implemented using a finite impulse response filter to ensure consistent filtering standards between the preparation and explosive phase data, reducing calculation errors.

[0126] Preferably, based on the degree of matching between the kinetic energy distribution ratio and the preset target interval, attention concentration evaluation results for multi-mode motion are generated, including:

[0127] Based on the current sports mode number Retrieve the corresponding preset target proportion center parameter from the system preset parameters. and preset allowable scale parameters ;

[0128] Calculate single-action attention score using Gaussian mapping function The calculation formula is:

[0129]

[0130] in The proportion of the kinetic energy distribution. It is an exponential function with the natural constant as its base;

[0131] According to the preset training plan sequence, the attention scores for each repetitive action are weighted and averaged to obtain a comprehensive training score. The calculation formula is:

[0132]

[0133] in For the total number of patterns, These are the preset weighting coefficients for the corresponding modes. This represents the average score across all trials in this model.

[0134] The comprehensive training score Compared with the preset evaluation threshold The comparison is performed, and the attention concentration evaluation results of the multi-modal motion are output. The logic is as follows:

[0135]

[0136] The sports mode number is a unique identifier used to distinguish different sports types. It is preferably 1 to 10, corresponding to common multi-mode sports such as throwing, basketball shooting, golf putting, and squatting, to cover mainstream training scenarios and facilitate the system to retrieve the corresponding preset parameters.

[0137] The preset target proportion center is the ideal value of the kinetic energy distribution proportion set for each sports mode. The preferred values ​​are 0.25 for throwing, 0.30 for basketball shooting, 0.35 for golf putting, and 0.20 for squatting, in order to adapt to the mechanical proportion characteristics of the attention preparation period in different sports modes through a large number of sample tests.

[0138] The preset tolerance scale is a reasonable range within which the proportion of kinetic energy distribution is allowed to deviate from the center of the preset target proportion. It is preferably between 0.05 and 0.08. The more precise the movement in the motion mode, the smaller the tolerance scale, in order to balance the rigor and practicality of the scoring and avoid small deviations from causing large fluctuations in the score.

[0139] The single-action focus score maps the proportion of kinetic energy distribution in a single action to an intuitive score reflecting the degree of attentional concentration. Calculated using a Gaussian mapping function, the score ranges from 0 to 1, with higher scores indicating that the attentional preparation for a single action is closer to the ideal state.

[0140] The preset training plan mode sequence is the order of movement modes included in a pre-defined training process. Preferably, it is a sequence that increases in intensity or complexity of movement to conform to conventional training logic and facilitate the system to load the corresponding parameters in sequence.

[0141] The total number of modes is the number of different types of exercise modes included in the preset training plan's mode sequence. It is preferably between 1 and 10 to balance training diversity and evaluation efficiency, and to avoid excessively long evaluation cycles due to too many modes.

[0142] The preset weighting coefficient for each exercise mode represents its proportion in the overall training score. Ideally, all modes should have equal weights, i.e., the weight of each mode should be the reciprocal of the total number of modes. This ensures a balanced contribution of each exercise mode to the final evaluation result and avoids a single mode dominating the score.

[0143] The pattern trial score mean is the arithmetic mean of the attention scores for each repeated action in a specific movement pattern. It reflects the overall level of user concentration in that pattern.

[0144] The overall training score is the result of a weighted sum of the average scores from all trial sessions across all exercise modes, weighted by preset coefficients. It comprehensively reflects the user's level of concentration throughout the entire training program.

[0145] The evaluation threshold is the critical score for judging whether attention is focused. A value of 0.65 is preferred, as it serves as a reasonable dividing line between focused and unfocused attention based on extensive user test data.

[0146] The attention concentration assessment result is the final judgment based on a comparison between the comprehensive training score and the assessment pass threshold. It is divided into two categories: focused attention and inattentive attention, providing users with intuitive training feedback.

[0147] Based on the sports mode number, a specific preset target proportion center and preset allowable scale are retrieved. The kinetic energy distribution proportion is converted into a single-action focus score using a Gaussian mapping function. Then, a weighted average is calculated using preset weighting coefficients to obtain a comprehensive training score. Finally, the result is output based on a fixed threshold. Conventional scoring methods often use uniform target standards or linear scoring, which cannot adapt to the differences between various sports modes. For example, the focus preparation period is higher in golf putting, so the preset target proportion center is set at 0.35, while the focus preparation period is lower in squats, set at 0.20. By using a specific target range and Gaussian mapping, the actual effect of focus preparation in different sports can be accurately reflected, avoiding scoring bias caused by uniform standards.

[0148] The preset target weighting and allowable scales for different sports modes include: explosive sports such as throwing (0.25±0.05, basketball shooting (0.30±0.06)); precision control sports such as golf putting (0.35±0.05); and strength sports such as squatting (0.20±0.08). The preset weighting coefficients for all modes are equal by default. If a particular type of sports needs to be emphasized, its weight can be increased to 1.2 times that of other modes, while maintaining a total weight of 1. The evaluation pass threshold of 0.65 is determined by testing 50 users with different skill levels and using the lowest score from the attention concentration group as the cutoff value. The preset training plan mode sequences are input manually through the device interface or by importing training files. The system completes parameter loading within 50 milliseconds when switching modes to ensure a smooth evaluation process.

[0149] It is important to note that all input data described in this solution is acquired in real-time through legal and compliant hardware interfaces with the user's full knowledge, explicit consent, and active cooperation. The preset parameters, prior constants, and statistical means are all derived from publicly available scientific literature data, de-identified general research datasets, or calibration data from laboratory environments, and do not contain any unauthorized sensitive third-party information. The system's data processing is limited to local or volatile memory computation transmitted via encrypted channels. There is no illegal collection, theft, or retention of user biometric data or infringement of user privacy without the user's knowledge. All parameter calls and generation comply with the principles of data minimization, legality, legitimacy, and necessity.

[0150] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A cognitive judgment training device for multimodal motion, characterized in that, The device includes a central monitoring unit worn on the user's sternum, limb sensing units worn on the user's limbs, and a data processing unit. The central monitoring unit is used to establish a human torso reference system with the chest as the reference point and to lock the gravity reference direction. The data processing unit is configured to perform the following steps: Based on the limb distribution structure parameters of the limb sensing unit in the human torso reference frame as the lever arm, and combined with the relative motion acceleration of the limb sensing unit relative to the central monitoring unit, a limb-torso related torque signal characterizing the body feedforward posture control is generated. Based on the temporal fluctuation of the associated torque signal, the main action trigger time is locked, and within a preset preparation time window before the main action trigger time, the expected cumulative amount of posture adjustment of the associated torque signal on the horizontal posture reference plane perpendicular to the gravity reference direction is calculated. The proportion of the expected cumulative posture adjustment in the total dynamics of the entire process, including the burst phase of the movement, is calculated. Based on the degree of matching between the proportion of kinetic energy distribution and the preset target interval, the attention concentration evaluation results of multi-mode movement are generated.

2. The cognitive judgment training device for multimodal motion according to claim 1, characterized in that, The central monitoring unit worn on the user's sternum, the limb sensing units worn on the user's extremities, and the data processing unit include: The central monitoring unit is worn above the xiphoid process of the user's sternum and serves as the origin of the human torso reference system. There are four limb sensing units, which are worn on the back of the user's left wrist joint, the back of the right wrist joint, the outside of the left ankle joint, and the outside of the right ankle joint, respectively, forming a distributed array centered on the central monitoring unit.

3. The cognitive judgment training device for multimodal motion according to claim 1, characterized in that, Establish a human torso reference frame with the chest as the reference point, and lock the gravity reference direction, including: The user is instructed to remain in a static standing posture as an initialization period. During this initialization period, the central monitoring unit collects triaxial acceleration data and calculates the corresponding arithmetic mean. The normalized vector of the arithmetic mean is used as the gravity reference direction, and the direction opposite to the gravity reference direction is used as the vertical axis direction of the human torso reference system. The forward axis direction of the human torso reference system is determined according to the preset wearing orientation of the central monitoring unit; The transverse axis direction of the human torso reference system is determined by the vector cross product operation between the forward axis direction and the vertical axis direction, thereby constructing an orthogonal human torso reference system.

4. The cognitive judgment training device for multimodal motion according to claim 1, characterized in that, Based on the limb distribution structure parameters of the limb sensing unit in the human torso reference frame as the lever arm, and combined with the relative motion acceleration of the limb sensing unit relative to the central monitoring unit, a limb-torso associated torque signal characterizing body feedforward posture control is generated, including: The spatial position vector of the limb sensing unit relative to the central monitoring unit is obtained when the user is in a standard upright initial posture, and the spatial position vector is used as the initial lever arm vector in the human torso reference frame. The user's weight data is acquired, and the limb segment mass corresponding to each limb sensing unit is calculated using a preset limb segment mass ratio coefficient. The real-time acceleration collected by the limb sensing unit is mapped to the human torso reference frame through an attitude rotation matrix, and the acceleration of the central monitoring unit in the human torso reference frame is subtracted to obtain the relative motion acceleration. For each limb sensing unit, the cross product of the initial lever arm vector and the relative motion acceleration is calculated, and multiplied by the corresponding limb segment mass to obtain the single limb torque component. The single-limb torque components of all the limb sensing units are vector-summed to generate the limb-torso related torque signal.

5. The multimodal motion cognitive judgment training device according to claim 4, characterized in that, Based on the timing fluctuations of the associated torque signal, the trigger time of the main action is locked, including: Construct a horizontal posture reference plane projection matrix perpendicular to the gravity reference direction, and project the limb-torso associated torque signal onto the horizontal posture reference plane to obtain the horizontal projection torque vector; Calculate the magnitude of the horizontal projected moment vector to obtain the real-time start-up intensity signal; The arithmetic mean and standard deviation of the background noise are calculated using the real-time start-up intensity signal during the initialization period; The activation threshold is constructed based on the arithmetic mean and standard deviation of the background noise and a preset threshold coefficient. The real-time activation intensity signal is scanned in chronological order, and the moment when the signal first continuously exceeds the activation determination threshold and the duration reaches a preset critical duration is locked as the main action trigger moment.

6. The cognitive judgment training device for multimodal motion according to claim 5, characterized in that, Within a preset preparation window prior to the main action triggering moment, the expected cumulative attitude adjustment of the associated torque signal on a horizontal attitude reference plane perpendicular to the gravity reference direction is calculated, including: Obtain the preset preparation window duration parameter, and determine the number of discrete sampling points corresponding to the preparation window duration parameter in combination with the system sampling frequency; Starting from the moment when the main action is triggered, backtracking in the negative direction of the time axis, select multiple discrete moments that cover the duration parameter of the preparation window; Obtain the magnitude of the horizontal projection moment vector corresponding to the multiple discrete moments, and sum all the magnitudes. Multiplying the summation result by the reciprocal of the system sampling frequency yields the expected cumulative posture adjustment amount.

7. The cognitive judgment training device for multimodal motion according to claim 6, characterized in that, Calculating the proportion of the expected cumulative posture adjustment in the total kinetic energy distribution of the entire process, including the burst phase of the movement, includes: Obtain the pre-set execution window duration parameter, and determine the number of execution sampling points corresponding to the execution window duration parameter in combination with the system sampling frequency; Starting from the moment when the main action is triggered, proceed in the positive direction of the time axis and select multiple discrete moments that cover the duration parameter of the execution window; Calculate the integral sum of the magnitudes of the horizontal projected moment vectors corresponding to the multiple discrete moments to obtain the dynamic quantity during the burst phase of the action; The total dynamic amount of the entire process is obtained by adding the cumulative amount of the expected posture adjustment to the dynamic amount of the burst phase of the action. The ratio of the expected cumulative attitude adjustment to the total dynamics of the entire process is calculated to obtain the proportion of kinetic energy distribution.

8. The cognitive judgment training device for multimodal motion according to claim 7, characterized in that, Based on the degree of matching between the kinetic energy distribution ratio and the preset target interval, attention concentration evaluation results for multi-modal motion are generated, including: Based on the current motion mode, retrieve the corresponding preset target proportion center and preset allowable scale from the system preset parameters; Calculate the square of the difference between the kinetic energy distribution ratio and the center of the preset target ratio, divide it by twice the square of the preset allowable scale parameter, and obtain the calculation result; take the negative number of the calculation result and perform exponential operation with the natural constant as the base to obtain the single action focus score; According to the preset training plan mode sequence, the single action focus score obtained from multiple repetitions of the action is weighted and averaged to obtain the comprehensive training score. The comprehensive training score is compared with a preset evaluation pass threshold: if the comprehensive training score is greater than or equal to the evaluation pass threshold, the result of attention concentration is output; if the comprehensive training score is less than the evaluation pass threshold, the result of attention inattention is output.

Images