Intelligent dumbbell counting method, medium and system

Abstract

This invention provides a method, medium, and system for counting repetitions with intelligent dumbbells. The method includes: real-time acquisition of raw sampling point data via a six-axis sensor; real-time preprocessing of the raw sampling point data to obtain preprocessed sampling point data; if the acceleration modulus values ​​corresponding to a consecutive preset first number of sampling point data are all greater than a preset trigger threshold, then training is initiated, and an adaptive threshold process and a motion posture recognition process are executed simultaneously; based on the sampling point data after training is initiated and the current trigger acceleration and trigger angular velocity thresholds, a counting process is executed. During this process, if the motion posture recognition process determines that the dumbbell's movement type has changed, then the adaptive threshold process is triggered. The core of the method is an adaptive threshold process, which dynamically calibrates the counting threshold based on the user's peak acceleration, angular velocity, and movement type, accurately adapting to the user's current personal exercise habits and improving counting accuracy.

Description

Technical Field

[0001] This invention relates to the field of intelligent fitness equipment technology, and in particular to intelligent dumbbell counting methods, media and systems. Background Technology

[0002] Currently, dumbbell repetition counters on the market mainly fall into three categories: manual counters rely on subjective operation, making it easy to forget counts and resulting in inaccurate counting; mechanical sensor counters have complex structures that are prone to wear and tear, are only compatible with a single movement, and have poor versatility; single-sensor counters mostly rely on accelerometers to judge repetitions based on fixed thresholds, which are simple in structure and low in cost, but have extremely low counting accuracy. A very small number have introduced six-axis sensors to improve counting accuracy, but they still have significant technical defects: First, they lack versatility, using fixed thresholds to judge repetitions, only adapting to specific training movements, and failing to adapt to the movement trajectories and data characteristics of different movements, making them prone to counting failures or errors in movements such as lateral raises and presses; second, they have weak anti-interference capabilities, making it difficult to accurately distinguish invalid movements caused by interference signals such as hand tremors, accidental touches, collisions, and environmental vibrations, which easily lead to miscounting; third, the overall power consumption of the devices is high, and the battery life is weak, making it impossible to support users for long training sessions.

[0003] Therefore, there is an urgent need for a completely new intelligent dumbbell counting method, medium, and system to solve the above-mentioned technical problems simultaneously. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an intelligent dumbbell counting method, medium and system that significantly improves the accuracy, adaptability and practicality of dumbbell counting.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: Methods for counting repetitions with smart dumbbells include: S1: Real-time acquisition of raw sampling point data, including triaxial acceleration and triaxial angular velocity, via a six-axis sensor; S2: Perform real-time preprocessing on the original sampling point data to obtain preprocessed sampling point data; S3: If the acceleration modulus values ​​corresponding to the first number of consecutive sampling points are all greater than the preset start threshold, then training is started, and the adaptive threshold process, motion posture recognition process, and count process are executed sequentially. The adaptive threshold process includes: Acquire sampling point data corresponding to a second consecutive preset number of training actions; based on the sampling point data of each training action, extract its acceleration peak value and angular velocity peak value, and calculate the average value of the acceleration peak value and the average value of the angular velocity peak value of the second consecutive preset number of training actions; based on the acquired sampling point data, identify the action type by the axial proportion of the gyroscope; based on the average value of the acceleration peak value and the average value of the angular velocity peak value, and the identified action type, calibrate and update the trigger count acceleration threshold and the trigger count angular velocity threshold; The motion posture recognition process includes: The system acquires a third consecutive set of sampling point data, calculates the corresponding real-time attitude angle change trajectory data of the dumbbell, and constructs the corresponding action feature vector. If the constructed action feature vector matches the action feature vector of a certain action type in the typical action feature library, then the action type is determined to be the current action type of the dumbbell. S4: During the period when the counting process is executed based on the current trigger counting acceleration threshold and trigger counting angular velocity threshold, starting from the continuously preset first number of sampling point data, if the motion posture recognition process determines that the dumbbell's motion type has changed, then the adaptive threshold process is triggered.

[0006] Optionally, the counting process includes: The system acquires and judges sampling point data starting from the first consecutive preset number of sampling point data in real time. If the acceleration magnitude and angular velocity magnitude corresponding to multiple consecutive sampling point data are both greater than the corresponding trigger count acceleration threshold and trigger count angular velocity threshold, and the corresponding time span exceeds the preset duration, and the corresponding dumbbell real-time posture angle change trajectory data conforms to the action characteristics of the action type currently determined by the motion posture recognition process, then a valid action is determined to be identified, and the count result is updated by 1.

[0007] Optionally, in the adaptive thresholding process, obtaining sampling point data corresponding to a consecutive preset second number of training actions includes: The real-time acquired sampling point data are identified sequentially, and all sampling point data during the period from the peak value of acceleration magnitude falling back to the start threshold are extracted to form the sampling point data corresponding to a single training action; Acquire sampling point data corresponding to a preset second number of consecutive training actions.

[0008] Optionally, in the adaptive thresholding process, the step of identifying the action type based on the acquired sampling point data and the gyroscope axial proportion includes: If the proportion of Z-axis angular velocity in the acquired sampling data exceeds the preset proportion, it is identified as a bicep curl movement. If the proportion of X-axis angular velocity in the acquired sampling data exceeds a preset ratio, it is identified as a lateral lift-type movement.

[0009] Optionally, in the adaptive thresholding process, calibrating and updating the trigger-count acceleration threshold and trigger-count angular velocity threshold based on the average peak acceleration and average peak angular velocity, and the identified action type, includes: If the exercise is identified as a bicep curl, the trigger count acceleration threshold is calibrated to the first coefficient * the average value of the peak acceleration, and the trigger count angular velocity threshold is calibrated to the second coefficient * the average value of the peak angular velocity. If the movement is identified as a lateral lift, the trigger acceleration threshold is calibrated to the second coefficient * the average value of the peak acceleration, and the trigger angular velocity threshold is calibrated to the first coefficient * the average value of the peak angular velocity; wherein the value of the first coefficient is greater than the value of the second coefficient. The system updates the stored trigger count acceleration threshold and trigger count angular velocity threshold based on the calibrated trigger count acceleration threshold and trigger count angular velocity threshold.

[0010] Optionally, in the motion posture recognition process, the calculation of the corresponding real-time dumbbell posture angle change trajectory data and the construction of the corresponding motion feature vector include: Quaternion attitude calculation is used to calculate the sampled point data obtained in this case to obtain the real-time attitude angle change trajectory data of the dumbbell. The real-time attitude angle change trajectory data of the dumbbell includes pitch angle data, roll angle data and yaw angle data. Based on the sampling point data and the real-time attitude angle change trajectory data of the dumbbell obtained this time, a corresponding motion feature vector is constructed. The motion feature vector includes three dimensions: attitude angle change range, angular velocity direction distribution, and acceleration magnitude change trend.

[0011] Optionally, the step of constructing a corresponding motion feature vector based on the sampled point data and the real-time dumbbell posture angle change trajectory data obtained this time includes: Based on the real-time attitude angle change trajectory data of the dumbbell, the pitch angle change range = maximum pitch angle - minimum pitch angle and the roll angle change range = maximum roll angle - minimum roll angle were calculated respectively. The larger value between the pitch angle change range and the roll angle change range is selected and marked as the quantized value of the attitude angle change range dimension; Based on the sampling data obtained this time, calculate the cumulative sum of the three-axis angular velocities; Calculate the proportion of the Z-axis angular velocity and the X-axis angular velocity in the sum of the three-axis angular velocities, and select the larger value as the dimensional quantization value of the angular velocity direction distribution dimension. Based on the sampling point data obtained this time, the corresponding acceleration magnitude data is calculated; Based on the acceleration peak value in the acceleration modulus data, the sampling point data obtained this time is divided into stages, and the sampling point data corresponding to the rising segment, peak segment and falling segment are obtained respectively. The sampling point data corresponding to the rising segment, peak segment, and falling segment are respectively fitted with parameters to obtain the slope of the rising segment, the peak value, and the slope of the falling segment, and these are marked as the dimensional quantization value of the acceleration modulus change trend.

[0012] Optionally, the preprocessing in S2 includes: The triaxial acceleration and triaxial angular velocity in the original sampling point data are smoothed by the moving average filtering algorithm to obtain the filtered triaxial acceleration and filtered triaxial angular velocity. Based on the zero-drift reference value of the gyroscope in the six-axis sensor, the filtered three-axis angular velocity is calibrated to obtain the calibrated three-axis angular velocity.

[0013] Another technical solution provided by this invention is: A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, enables the implementation of the above-described intelligent dumbbell counting method.

[0014] Another technical solution provided by this invention is: A smart dumbbell counting system includes the aforementioned computer-readable storage medium and a processor; the processor is disposed inside the dumbbell body and / or inside a smart terminal, and is capable of calling and executing the computer program stored in the computer-readable storage medium.

[0015] The beneficial effects of this invention are as follows: Based on a six-axis sensor, this invention collects dumbbell motion data and determines training initiation by continuously sampling multiple acceleration modulus values, effectively avoiding false triggers and improving the reliability of motion recognition. Through an adaptive threshold process, it distinguishes different motion types based on the average peak values ​​of acceleration and angular velocity from multiple consecutive sets of user movements, combined with the gyroscope's axial proportion. Based on this data, it calibrates the counting threshold in real-time and specifically, accurately adapting to the user's personalized exercise habits (motion amplitude, motion intensity), solving the problem of insufficient adaptability of traditional fixed thresholds. Through a motion posture recognition process, it constructs a multi-dimensional motion feature vector for the current training motion in real-time and matches it with a typical library, achieving accurate motion type classification with high recognition accuracy and strong anti-interference capabilities. Through a counting process, it comprehensively determines valid movements by combining personalized dual thresholds, duration, and motion type stability, ensuring accurate and reliable counting. Especially when switching motion types, it can automatically recalibrate the counting threshold, achieving full-process adaptive adjustment, accommodating the counting needs of various fitness motion types, and significantly improving the accuracy, adaptability, and practicality of dumbbell counting. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a smart dumbbell counting method provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the adaptive threshold process A described in this embodiment of the invention; Figure 3 This is a flowchart illustrating motion posture recognition process B as described in this embodiment of the invention. Figure 4 This is a flowchart illustrating the counting process C described in this embodiment of the invention. Detailed Implementation

[0017] To explain in detail the technical principles, specific implementable solutions, possible application scenarios, and achievable objectives and effects of the present invention, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. It is understood that the embodiments described herein and the embodiments shown in the accompanying drawings are only used to more clearly illustrate the technical solutions of the present invention, and are therefore only examples intended to explain the present invention, and should not be construed as limiting the present invention. It should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0018] Please see Figures 1 to 4 This invention provides a method for counting dumbbell repetitions.

[0019] like Figure 1 As shown, an embodiment of the present invention provides a method for counting dumbbell repetitions, comprising the following steps S1-S4: S1: Real-time acquisition of raw sampling point data, including triaxial acceleration and triaxial angular velocity, via a six-axis sensor.

[0020] The six-axis sensor (6DoF / IMU) is an inertial measurement unit integrating a three-axis accelerometer and a three-axis gyroscope. It can simultaneously output linear acceleration and angular velocity in three-dimensional space, and achieve precise attitude, motion, and orientation detection through a fusion algorithm. In this embodiment, the six-axis sensor can acquire the three-axis acceleration (Ax, Ay, Az) and three-axis angular velocity (Gx, Gy, Gz) of the dumbbell in real time.

[0021] S2: Perform real-time preprocessing on the original sampling point data to obtain preprocessed sampling point data, thereby eliminating noise interference and data drift.

[0022] In some specific implementations, the preprocessing procedure includes the following sub-steps: 2-1. The moving average filtering algorithm is used to smooth the triaxial acceleration and triaxial angular velocity in the original sampling point data to obtain the filtered triaxial acceleration and filtered triaxial angular velocity.

[0023] Here, the original sampled data is processed by sliding smoothing filter to filter out the high-frequency noise caused by the slight hand tremors and environmental vibrations when the user operates the dumbbell.

[0024] For example, the sliding window size is set to 3-15 sampling points; the following example uses 5 sampling points. The raw sampling point data received in real time is smoothed using the sliding window using the following formula: ;in, For the original data of the i-th sampling point, This represents the filtered data for the i-th sampling point.

[0025] 2-2. Based on the zero-drift reference value of the gyroscope in the six-axis sensor, the filtered three-axis angular velocity is calibrated to obtain the calibrated three-axis angular velocity.

[0026] Here, zero-drift calibration is performed on the gyroscope data to eliminate attitude calculation errors caused by the zero drift of the six-axis sensor itself.

[0027] For example, the zero-drift reference value of the gyroscope is obtained by collecting gyroscope data from dozens or hundreds (preferably 100) sampling points of the dumbbell device in a stationary state, namely the three-axis angular velocities (Gx, Gy, Gz); then the average value is calculated as the zero-drift reference value G. static .

[0028] For example, zero-drift calibration is performed on the filtered triaxial angular velocities received in real time according to the following formula: Among them, G static This is the average value of the gyroscope data in a static state, i.e., the zero-drift reference value; G cal (n) represents the calibrated triaxial angular velocity data.

[0029] Optionally, the zero-drift calibration described above can be performed after the sliding smoothing filter processing in 2-1 above, and correspondingly, G filter (n) represents the filtered triaxial angular velocity; it can also be performed afterward, i.e., first performing zero-drift calibration on the raw sampled data received in real time, and then performing sliding smoothing filtering. Accordingly, G filter (n) represents the triaxial angular velocity in the original sampling point data. The "triaxial angular velocity data" in the sliding smoothing filter process in 2-1 above becomes the "calibrated triaxial angular velocity data".

[0030] 2-3. Obtain the preprocessed sampling point data consisting of the filtered triaxial acceleration and the calibrated triaxial angular velocity.

[0031] It is understood that the preprocessed sampling point data refers to the triaxial acceleration and triaxial angular velocity, which have been processed by sliding smoothing filtering, and the triaxial angular velocity has been processed by zero drift calibration.

[0032] S3: Training status assessment.

[0033] First, the dumbbell system determines whether the user has entered a training state based on the preprocessed sampling data received in real time. Specifically, it determines the acceleration magnitude A corresponding to a first preset number K1 of sampling data points. amg If all values ​​are greater than the preset start threshold, then training is started; otherwise, training is not started.

[0034] Wherein, the acceleration modulus The selectable value range of the first quantity K1 is 3-15; preferably, the first quantity K1=5. The selectable value range of the activation threshold is 0.2-1g, preferably 0.5g.

[0035] Secondly, if training is initiated, the adaptive threshold process A, motion pose recognition process B, and count process C are executed sequentially.

[0036] Here, the adaptive threshold process A, motion posture recognition process B, and count process C are all executed based on the sampling point data obtained in real time after the training starts, that is, the sampling point data including the first number K1 sampling point data obtained in real time from the above S2.

[0037] In some specific implementations, a data pool is pre-set to store the target processing data of process AC. After determining the start of training, the sampling point data of the first number K1 and all the preprocessed sampling point data obtained in real time through S2 thereafter are stored in the data pool in the sampling order; the execution of the adaptive threshold process A, motion posture recognition process B and counting process C are all based on the data pool, and the sampling point data is obtained from it in sequence.

[0038] In some specific implementations, if it is determined that training has not started, a low-power mode is entered.

[0039] Optionally, the low-power mode includes reducing the sampling frequency of the six-axis sensor and controlling the MCU to sleep, in order to save dumbbell power consumption to the maximum extent.

[0040] In this embodiment, the adaptive threshold process A automatically calibrates the count threshold based on the average peak values ​​of acceleration and angular velocity of the first few training movements performed by the current user with the dumbbell (reflecting their amplitude and intensity of movement), combined with the movement type distinguished by the gyroscope axis proportion, so as to accurately adapt to the current user's personalized exercise habits (movement type, amplitude, and intensity).

[0041] Specifically, such as Figure 2 As shown, the implementation process of the adaptive threshold procedure A includes the following sub-steps: A1: Obtain the sampling point data corresponding to the second preset number of training actions.

[0042] Optionally, the second quantity K2 can be selected in the range of 3-15; preferably, the second quantity K2=3.

[0043] For example, after determining that training has started, the sampling point data of the first 3 complete training actions are recorded immediately.

[0044] In some specific implementations, the identification and acquisition of sampling point data corresponding to a single complete training action are determined based on the following rule: during the movement of the training action from the "starting position" to the "ending position," the acceleration magnitude shows a change characteristic of rising from near the initial state threshold (i.e., the aforementioned start threshold) to a peak value, and then falling back to the initial state threshold. Based on this change rule, the sampling data segment corresponding to a single complete training action can be accurately identified and selected.

[0045] As a preferred embodiment, the process for identifying and acquiring the sampling point data corresponding to a single complete training action is as follows: After determining the start of training, sampling point data are sequentially retrieved from the data pool and identified in order (retrieval sequence), extracting the acceleration magnitude A. amgThe sampling point data segment from the peak drop to the start threshold constitutes the sampling point data corresponding to a single training action; accordingly, the sampling point data corresponding to each of the three consecutive training actions is obtained.

[0046] A2: Based on the sampling point data corresponding to each training action, extract its peak acceleration A. peak and peak angular velocity G peak ; A3: Based on the peak acceleration A corresponding to each training movement peak and peak angular velocity G peak The average peak acceleration was calculated. and peak and average angular velocity ; A4: Based on the sampling point data obtained in A1, the action type is identified by the axial proportion of the gyroscope; In some specific implementations, the process of identifying the action type by the gyroscope's axial proportion includes: If the proportion of Z-axis angular velocity (perpendicular to the ground) in the acquired sampling data exceeds the preset proportion, it is identified as a curling exercise, that is, the training exercise performed by the user with the dumbbell is a curling exercise. If the proportion of X-axis angular velocity (horizontal direction) in the acquired sampling point data exceeds the preset proportion, it is identified as a lateral raise exercise, that is, the training exercise performed by the user with the dumbbell is a lateral raise exercise.

[0047] The selectable value range for the preset ratio is 60-80%; preferably 70%.

[0048] A5: Average peak acceleration calculated based on A3 and peak and average angular velocity And the action type identified by A4, and the trigger count acceleration threshold Th A and trigger count angular velocity threshold Th G Perform calibration and update.

[0049] In some specific implementations, the A5 step is specifically implemented as follows: If A4 is identified as a bicep curl, then the acceleration threshold Th will be triggered. A The calibration is based on the first coefficient P1*, which represents the average peak acceleration value. This will trigger the angular velocity threshold Th. G The calibration is based on the second coefficient P2*, which represents the peak value of the angular velocity. ; If the movement is identified as a lateral raise, the acceleration threshold Th will be triggered. A The calibration is based on the second coefficient P2*, which represents the peak-to-average acceleration. This will trigger the angular velocity threshold Th. G The calibration is based on the first coefficient P1 * the average value of the peak angular velocity. ; Update the system-stored trigger count acceleration threshold and trigger count angular velocity threshold to the calibrated trigger count acceleration threshold Th. A and trigger count angular velocity threshold Th G Optionally, the trigger acceleration threshold and trigger angular velocity threshold are stored in the system flash, and the subsequent counting determination (i.e., counting process C) will be based on this.

[0050] The value of the first coefficient P1 is greater than the value of the second coefficient P2. The selectable values ​​of the first coefficient P1 and the second coefficient P2 are in the range of 0.3-0.9.

[0051] For example, if the Z-axis angular velocity (perpendicular to the ground) of the gyroscope accounts for more than 70% during the three movements (determined as a curl-type movement), then a trigger acceleration threshold Th is set. A =0.6* Trigger count angular velocity threshold Th G =0.5* If the X-axis angular velocity (horizontal direction) of the gyroscope accounts for more than 70% (determined as a lateral lift motion), then set the trigger count acceleration threshold Th. A =0.5* Trigger count angular velocity threshold Th G =0.6* (Because the acceleration amplitude of the lateral raise is relatively small, but the rotational posture is obvious).

[0052] In this embodiment, the motion posture recognition process B, based on the sampling point data used to determine the start of training and the preprocessed sampling point data received in real time thereafter, realizes motion trajectory and posture recognition through multi-dimensional data fusion.

[0053] Specifically, such as Figure 3 As shown, the implementation process of motion posture recognition process B includes the following sub-steps: B1: For the sampling point data used to determine the start of training and the preprocessed sampling point data received in real time thereafter, sequentially acquire a third preset number K3 of sampling point data in order (receiving order), calculate the real-time attitude angle change trajectory data of the dumbbell corresponding to each acquired sampling point data, and construct the corresponding motion feature vector.

[0054] The third quantity K3 is determined based on the theoretical time it takes for a user to complete a full training exercise with dumbbells, combined with the sampling frequency. For example, if the theoretical time for a user to complete a full bicep curl is 0.8s and the sampling frequency is 63Hz, by setting the third quantity K3=50 (0.8s / 0.016ms≈50), the sampling point data obtained in a single session corresponds to the sampling point data for a full training exercise.

[0055] The selectable value range for the third quantity K3 is 20-200; preferably, the third quantity K3=20 or K3=50.

[0056] In some specific implementations, the data from a single sampling point are calculated to obtain the corresponding real-time attitude angle change trajectory data of the dumbbell (pitch angle θ, roll angle φ, yaw angle ψ), and the corresponding motion feature vector is constructed, specifically including: First, quaternion attitude calculation is used to calculate the real-time attitude angle change trajectory data of the dumbbell obtained from the sampled data. The real-time attitude angle change trajectory data of the dumbbell includes pitch angle data, roll angle data, and yaw angle data.

[0057] Secondly, based on the sampling point data and the real-time attitude angle change trajectory data of the dumbbell obtained this time, a corresponding motion feature vector is constructed. The motion feature vector includes three dimensions: attitude angle change range, angular velocity direction distribution, and acceleration magnitude change trend.

[0058] As a preferred embodiment, the calculation process for the real-time attitude angle change trajectory data of the dumbbell is as follows: Quaternion update equation: q = ½ q ω; where q is a quaternion (q0, q1, q2, q3), and ω is the angular velocity vector of the gyroscope's three axes (Gx, Gy, Gz); Attitude angle conversion: Pitch angle θ = arctan2(2(q0q2+q1q3), 1-2(q2) 2 +q3 2 )); Roll angle =arcsin(2(q0q1-q2q3)); Yaw angle Ψ = arctan2(2(q0q3+q1q2),1-2(q1) 2 +q3 2 )).

[0059] As a preferred embodiment, the process of constructing the action feature vector is as follows: (1) Range of attitude angle variation: Based on the real-time attitude angle change trajectory data of the dumbbell, the pitch angle change amplitude θ = maximum pitch angle - minimum pitch angle and the roll angle change amplitude φ = maximum roll angle - minimum roll angle were calculated respectively. The larger of the pitch angle change range and the roll angle change range is selected and marked as the quantized value of the attitude angle change range dimension.

[0060] (2) Angular velocity direction distribution: Based on the sampling data obtained this time, calculate the sum (absolute value) of the triaxial angular velocities; Calculate the proportion of the Z-axis angular velocity and the X-axis angular velocity in the sum of the three-axis angular velocities, and select the larger value as the dimensional quantization value of the angular velocity direction distribution dimension.

[0061] (3) Trend of acceleration magnitude variation: Based on the sampling point data obtained this time, the corresponding acceleration magnitude data is calculated; Based on the acceleration peak value in the acceleration modulus data, the sampling point data obtained this time is divided into stages, and the sampling point data corresponding to the rising segment, peak segment and falling segment are obtained respectively. The sampling point data corresponding to the rising segment, peak segment, and falling segment are respectively fitted with parameters to obtain the slope of the rising segment, the peak value, and the slope of the falling segment, and these are marked as the dimensional quantization value of the acceleration modulus change trend.

[0062] For example, taking a single standard bend as an example, the process of constructing its corresponding action feature vector is as follows: 1. Data processing based on a six-axis sensor: The sampling frequency was 63Hz (one set of data was collected every 16ms). A single complete bicep curl lasted 0.8s, and a total of 50 sampling points were collected (0.8s / 0.016ms≈50). Processed data: filtered triaxial acceleration data (Ax, Ay, Az), calibrated and filtered triaxial angular velocity data (Gx, Gy, Gz), and calculated attitude angle change trajectory data (pitch angle θ, roll angle φ, yaw angle ψ).

[0063] 2. Attitude angle variation range: Pitch angle θ data within the motion cycle: [10°, 15°, 22°, ..., 85°, 80°, 75°, ..., 12°], a total of 50 points; Calculate: Max(θ) = 85°, Min(θ) = 10°, the range of change = 85° - 10° = 75°; Roll angle φ data: [3°, 4°, ..., 5°], variation range = 2° (negligible, non-dominant dimension); The final quantization value for this dimension is 75° (only the range of variation of the dominant dimension is retained to simplify the calculation and highlight the core features).

[0064] 3. Angular velocity direction distribution: The sum of the three-axis angular velocities (absolute value) within the motion cycle: Z-axis (perpendicular to the ground, the dominant axis of rotation in bicep curl): Σ|Gz|=14200° / s, which is the sum of the Z-axis sampled data from each of the three-axis angular velocities; for example, Gz is the sum of 280° / s, 260° / s, etc. X-axis (horizontal direction): Σ|Gx|=850° / s; Y-axis (vertical direction): Σ|Gy|=620° / s; Calculate the percentage: Z-axis percentage = 14200 / (14200+850+620)×100%≈89.2%; X-axis percentage ≈ 5.3%; Y-axis percentage ≈ 3.9%; Ultimately, the quantification value for this dimension was 89.2% (Z-axis dominant, accounting for more than 85%, consistent with the characteristics of bicep curl).

[0065] 4. Trend of acceleration magnitude variation: The acceleration magnitude A corresponding to the sampling point data within the action cycle amg [1.0g (initial), 1.5g, 2.3g, ..., 3.2g (peak), 3.1g, ..., 1.1g (end)]; Phase division: Ascent phase: 1-20 sampling points (time 0-0.32s), acceleration magnitude A amg Increased from 1.0g to 3.2g; Peak range: 21-25 sampling points (time 0.336-0.4s), acceleration magnitude A amg Maintain at 3.0-3.2g; Descent phase: 26-50 sampling points (time 0.416-0.8s), acceleration magnitude A amg Reduced from 3.1g to 1.1g.

[0066] Fitting parameters: The slope of the ascending segment = (3.2g - 1.0g) / 0.32s ≈ 6.875g / s; Peak value = 3.2g; The slope of the descent segment = (1.1g - 3.1g) / 0.384s ≈ -5.208g / s; The final quantified values ​​for this dimension are (6.875g / s, 3.2g, -5.208g / s) (the combined parameters fully reflect the force change trend of the bicep curl).

[0067] B2: If the motion feature vector constructed this time matches the motion feature vector of a motion type in the typical motion feature library, then the motion type is determined to be the current motion type of the dumbbell.

[0068] The typical motion feature library pre-stores motion feature vectors corresponding to various motion types (bicep curl, lateral raise, press).

[0069] In some specific implementations, if the matching degree between the constructed action feature vector and the action feature vector of action type a in the typical action feature library exceeds a preset percentage, the current action type is determined to be action type a. Optionally, the preset percentage ranges from 70% to 90%; preferably, it is 85%.

[0070] In some other implementations, if a match is not found, it is determined to be a custom action, and the number of attempts is counted based on its own characteristics.

[0071] For example, the implementation process of making the count judgment based on its own characteristics is as follows: First, based on the real-time collected three-axis acceleration and three-axis angular velocity, as well as the corresponding calculated attitude angle data, the custom action's unique features such as peak acceleration, peak angular velocity, attitude angle variation range, and action duration are extracted. Secondly, using the user's first three complete actions as calibration samples, the personalized acceleration threshold, angular velocity threshold, attitude determination interval, and duration threshold for that action are calculated. Finally, subsequent actions must simultaneously meet the acceleration threshold, angular velocity threshold, attitude matching, and duration requirements to be considered valid actions and counted. Furthermore, the thresholds are dynamically updated during the counting process to improve counting accuracy.

[0072] S4: During the execution of the counting process, if the motion type of the dumbbell changes according to the motion posture recognition process, the adaptive threshold process is triggered.

[0073] In this embodiment, the counting process C is executed based on the sampling data used to start training and the sampling data acquired in real time thereafter, as well as the calibration results of the adaptive threshold process A, which can achieve accurate counting of the current user according to their personalized training habits (motor amplitude, speed).

[0074] Specifically, such as Figure 4 As shown, the implementation process of the counting process C includes the following sub-steps: C1: The sampling point data used to determine the start of training and the preprocessed sampling point data received in real time are analyzed sequentially (in the order of reception), and the following judgments are made based on the analysis results: (1) The acceleration magnitude A corresponding to multiple consecutive sampling points amg All are greater than the currently stored trigger-count acceleration threshold Th A ; (2) The angular velocity magnitude G corresponding to the data of the consecutive multiple sampling points amg All are greater than the currently stored trigger count angular velocity threshold value Th G Angular velocity magnitude ; (3) The time span corresponding to the data of the multiple consecutive sampling points exceeds the preset duration h; the selectable value range of the preset duration h is 0.2-3s, and the preferred value is 0.3; it can be understood that the duration of the action meets the standard; (4) The attitude angle changes of the data from the multiple consecutive sampling points conform to the action characteristics corresponding to the action type currently determined in the motion posture recognition process B. That is, the attitude angle changes of the current training action conform to the characteristics of the current action type.

[0075] For example, if the pitch angle θ changes more than 60° during a curl, it is considered the "start phase" of the movement. When the acceleration modulus Aamg drops below 0.3*ThA based on subsequent sampling data, and the attitude angle returns to near the initial attitude (change less than 10°), it is considered the "completion phase" of the movement. At this time, the count counter is incremented by 1, and the count result is stored in flash.

[0076] C2: If all of the above (1) to (4) are satisfied, then a valid action is identified and the count result is updated by 1.

[0077] It is understandable that the above counting process C is configured with an invalid action filtering mechanism. When only the acceleration or angular velocity exceeds the corresponding counting threshold, or when both dimensions briefly exceed the counting threshold but the attitude angle change is less than the preset angle or the action duration is shorter than the preset duration h, it is judged as an invalid action and the counting is rejected.

[0078] Specifically, the invalid action filtering mechanism excludes the following two cases and does not execute the count: 1. Single-dimensional data trigger threshold: If only one dimension of acceleration or angular velocity exceeds the corresponding trigger count threshold, and the other dimension does not meet the standard (for example, hand shaking only causes acceleration fluctuations, but angular velocity does not change significantly), it is judged as an interference action and will not be counted. 2. Incomplete or insufficiently continuous movements: Although the acceleration and angular velocity briefly exceed the corresponding trigger count threshold, if the range of attitude angle change is less than the threshold (e.g., 30°), or the duration of the movement is shorter than the preset duration h (e.g., due to an accidental collision with a dumbbell), it is judged as a non-training movement, directly filtered out, and not counted.

[0079] In this embodiment, if the type of exercise performed by the user using dumbbells changes, the adaptive threshold process A will be automatically initiated to recalibrate the repetition threshold according to the current user's training characteristics. This enables the repetition parameters to be adaptively adjusted throughout the training process based on the type of exercise and training habits (amplitude and intensity) of the current user's exercise, thus adapting to the training habits of different users and achieving accurate repetition counting. This significantly improves the accuracy, adaptability, and practicality of dumbbell repetition counting.

[0080] In this embodiment, the adaptive threshold process A is triggered immediately after the training is initiated, and is triggered again after the motion posture recognition process B determines that the action type has changed. The motion posture recognition process B is triggered after the first round of adaptive threshold process A, sequentially determining the current action type based on multi-sampling point data until the analysis of all sampling point data is completed, at which point training ends. The counting process C is initiated after the first round of motion posture recognition process B, starting the analysis and judgment of the sampling point data used to initiate training from scratch until training ends.

[0081] The following example, using a specific scenario of counting repetitions for a standard bicep curl, will help to better understand the smart dumbbell repetition counting method described in this embodiment: The standard dumbbell curl movement is characterized by: holding dumbbells with both hands hanging naturally, bending your elbows upwards to lift the dumbbells to your chest, and then slowly lowering them back to the starting position; during the movement, the dumbbells mainly move in the vertical direction, with the pitch angle θ varying from approximately 70° to 90°, and the acceleration mainly concentrated along the Y-axis (vertical direction). System processing procedure: After initialization, the 3-axis accelerometer collected Y-axis acceleration data that increased from 1g (gravitational acceleration at rest) to 3.2g (peak lift), and the 3-axis gyroscope collected Z-axis angular velocity peak value of 280° / s; Based on the calibration of the first N (N≤3) actions through the adaptive threshold process A, the trigger count acceleration threshold ThA=1.92g (0.6*3.2g) and the trigger count angular velocity threshold ThG=140° / s (0.5*280° / s) are adaptively set. The counting process starts from the first movement, which includes N movements in the adaptive threshold process A. Taking the fourth movement as an example, the Y-axis acceleration reaches 3.1g (exceeding ThA), the Z-axis angular velocity reaches 260° / s (exceeding ThG), the pitch angle θ changes from 10° to 85° (consistent with bicep curl characteristics), and the movement lasts for 0.8s. This movement is judged as a valid movement, and the count counter is incremented by 1. During training, the user's hand trembled slightly (Y-axis acceleration fluctuated to 1.8g, which did not reach ThA, so the system judged it as an invalid action and did not count it. The current cumulative count is 3).

[0082] Another embodiment of the present invention provides a computer-readable storage medium based on the above embodiments.

[0083] The computer-readable storage medium of this embodiment stores a computer program that, when executed by a processor, implements the intelligent dumbbell counting method described in the above embodiment. The steps included in the method will not be repeated here; please refer to the description in the above embodiment for details.

[0084] Those skilled in the art will understand that all or part of the processes in the above technical solutions can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the methods described above. After being executed by a processor, the program can also achieve the beneficial effects of the corresponding methods.

[0085] The storage medium can be a disk, optical disc, read-only memory (ROM), or random access memory (RAM), etc.

[0086] Another embodiment of the present invention provides an intelligent dumbbell counting system based on the above embodiments.

[0087] This embodiment provides an intelligent dumbbell counting system, including the computer-readable storage medium and processor described in the previous embodiment.

[0088] Specifically, the processor can be a MUC processor located inside the dumbbell body; or it can be a MUC processor located inside a smart terminal (such as a mobile phone, television, tablet, etc.). The processor can call and execute the computer program stored in the computer-readable storage medium described in the previous embodiment to implement the smart dumbbell counting method described in the above embodiment. The steps included in the method will not be described again here; please refer to the description in the above embodiments for details.

[0089] In summary, the intelligent dumbbell repetition counting method, medium, and system provided by this invention are based on dumbbell motion data collected by a six-axis sensor. The core of the system employs an adaptive threshold process, dynamically calibrating the repetition counting threshold based on the user's acceleration, peak angular velocity, and movement type. This allows for precise adaptation to the user's individual exercise habits (amplitude, intensity, and type of movement). Combined with posture recognition and multi-condition repetition counting, accurate and reliable counting is achieved. In particular, the automatic recalibration of the repetition counting threshold during movement type switching significantly improves the accuracy and adaptability of dumbbell repetition counting.

[0090] Although preferred embodiments of the invention have been described, those skilled in the art, upon understanding the basic inventive concept, can make other changes and modifications to the embodiments. Therefore, the appended claims are intended to cover the preferred embodiments and all equivalent modifications falling within the scope of protection defined by the claims and their equivalents. Clearly, those skilled in the art can make various alterations and variations to the invention without departing from its spirit and scope. If such modifications and variations fall within the scope of protection defined by the claims and their equivalents, the invention also intends to include them.

[0091] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware.

[0092] In this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "multiple" means two or more, unless otherwise explicitly specified. The term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone.

[0093] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0094] In this specification, the use of terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refers to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the present invention. The illustrative expressions of the above terms in this specification should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0095] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for counting repetitions with intelligent dumbbells, characterized in that, include: S1: Real-time acquisition of raw sampling point data, including triaxial acceleration and triaxial angular velocity, via a six-axis sensor; S2: Perform real-time preprocessing on the original sampling point data to obtain preprocessed sampling point data; S3: If the acceleration modulus values ​​corresponding to the first number of consecutive sampling points are all greater than the preset start threshold, then training is started, and the adaptive threshold process, motion posture recognition process, and count process are executed sequentially. The adaptive threshold process includes: Acquire sampling point data corresponding to a second consecutive preset number of training actions; based on the sampling point data of each training action, extract its acceleration peak value and angular velocity peak value, and calculate the average value of the acceleration peak value and the average value of the angular velocity peak value of the second consecutive preset number of training actions; based on the acquired sampling point data, identify the action type by the axial proportion of the gyroscope; based on the average value of the acceleration peak value and the average value of the angular velocity peak value, and the identified action type, calibrate and update the trigger count acceleration threshold and the trigger count angular velocity threshold; The motion posture recognition process includes: The system acquires a third consecutive set of sampling point data, calculates the corresponding real-time attitude angle change trajectory data of the dumbbell, and constructs the corresponding action feature vector. If the constructed action feature vector matches the action feature vector of a certain action type in the typical action feature library, then the action type is determined to be the current action type of the dumbbell. S4: During the period when the counting process is executed based on the current trigger counting acceleration threshold and trigger counting angular velocity threshold, starting from the continuously preset first number of sampling point data, if the motion posture recognition process determines that the dumbbell's motion type has changed, then the adaptive threshold process is triggered.

2. The intelligent dumbbell counting method as described in claim 1, characterized in that, The counting process includes: The system acquires and judges sampling point data starting from the first consecutive preset number of sampling point data in real time. If the acceleration magnitude and angular velocity magnitude corresponding to multiple consecutive sampling point data are both greater than the corresponding trigger count acceleration threshold and trigger count angular velocity threshold, and the corresponding time span exceeds the preset duration, and the corresponding dumbbell real-time posture angle change trajectory data conforms to the action characteristics of the action type currently determined by the motion posture recognition process, then a valid action is determined to be identified, and the count result is updated by 1.

3. The intelligent dumbbell counting method as described in claim 1, characterized in that, In the adaptive thresholding process, obtaining sampling point data corresponding to a consecutive preset second number of training actions includes: The real-time acquired sampling point data are identified sequentially, and all sampling point data during the period from the peak value of acceleration magnitude falling back to the start threshold are extracted to form the sampling point data corresponding to a single training action; Acquire sampling point data corresponding to a preset second number of consecutive training actions.

4. The intelligent dumbbell counting method as described in claim 1, characterized in that, In the adaptive thresholding process, the step of identifying the action type based on the acquired sampling point data and the gyroscope axial proportion includes: If the proportion of Z-axis angular velocity in the acquired sampling data exceeds the preset proportion, it is identified as a bicep curl movement. If the proportion of X-axis angular velocity in the acquired sampling data exceeds a preset ratio, it is identified as a lateral lift-type movement.

5. The intelligent dumbbell counting method as described in claim 4, characterized in that, In the adaptive thresholding process, the step of calibrating and updating the trigger count acceleration threshold and trigger count angular velocity threshold based on the average peak acceleration value, the average peak angular velocity value, and the identified action type includes: If the exercise is identified as a bicep curl, the trigger count acceleration threshold is calibrated to the first coefficient * the average value of the peak acceleration, and the trigger count angular velocity threshold is calibrated to the second coefficient * the average value of the peak angular velocity. If the movement is identified as a lateral lift, the trigger acceleration threshold is calibrated to the second coefficient * the average value of the peak acceleration, and the trigger angular velocity threshold is calibrated to the first coefficient * the average value of the peak angular velocity; wherein the value of the first coefficient is greater than the value of the second coefficient. The system updates the stored trigger count acceleration threshold and trigger count angular velocity threshold based on the calibrated trigger count acceleration threshold and trigger count angular velocity threshold.

6. The intelligent dumbbell counting method as described in claim 1, characterized in that, In the motion posture recognition process, the calculation of the corresponding real-time posture angle change trajectory data of the dumbbell and the construction of the corresponding motion feature vector include: Quaternion attitude calculation is used to calculate the sampled point data obtained in this case to obtain the real-time attitude angle change trajectory data of the dumbbell. The real-time attitude angle change trajectory data of the dumbbell includes pitch angle data, roll angle data and yaw angle data. Based on the sampling point data and the real-time attitude angle change trajectory data of the dumbbell obtained this time, a corresponding motion feature vector is constructed. The motion feature vector includes three dimensions: attitude angle change range, angular velocity direction distribution, and acceleration magnitude change trend.

7. The intelligent dumbbell counting method as described in claim 6, characterized in that, Based on the sampling point data and the real-time dumbbell posture angle change trajectory data obtained this time, a corresponding motion feature vector is constructed, including: Based on the real-time attitude angle change trajectory data of the dumbbell, the pitch angle change range = maximum pitch angle - minimum pitch angle and the roll angle change range = maximum roll angle - minimum roll angle were calculated respectively. The larger value between the pitch angle change range and the roll angle change range is selected and marked as the quantized value of the attitude angle change range dimension; Based on the sampling data obtained this time, calculate the cumulative sum of the three-axis angular velocities; Calculate the proportion of the Z-axis angular velocity and the X-axis angular velocity in the sum of the three-axis angular velocities, and select the larger value as the dimensional quantization value of the angular velocity direction distribution dimension. Based on the sampling point data obtained this time, the corresponding acceleration magnitude data is calculated; Based on the acceleration peak value in the acceleration modulus data, the sampling point data obtained this time is divided into stages, and the sampling point data corresponding to the rising segment, peak segment and falling segment are obtained respectively. The sampling point data corresponding to the rising segment, peak segment, and falling segment are respectively fitted with parameters to obtain the slope of the rising segment, the peak value, and the slope of the falling segment, and these are marked as the dimensional quantization value of the acceleration modulus change trend.

8. The intelligent dumbbell counting method as described in claim 1, characterized in that, The preprocessing in S2 includes: The triaxial acceleration and triaxial angular velocity in the original sampling point data are smoothed by the moving average filtering algorithm to obtain the filtered triaxial acceleration and filtered triaxial angular velocity. Based on the zero-drift reference value of the gyroscope in the six-axis sensor, the filtered three-axis angular velocity is calibrated to obtain the calibrated three-axis angular velocity.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it can implement the intelligent dumbbell counting method according to any one of claims 1 to 8.

10. A smart dumbbell counting system, characterized in that, The invention includes the computer-readable storage medium as described in claim 9 and a processor; the processor is disposed inside the dumbbell body and / or inside the smart terminal, and is capable of calling and executing the computer program stored in the computer-readable storage medium.

Images