Human waist posture recognition and force state evaluation method and system based on inertial sensor

Abstract

The present application relates to the technical field of waist posture recognition, in particular to a human waist posture recognition and force state evaluation method and system based on an inertial sensor, comprising the following steps: collecting original motion data in real time through an inertial measurement unit worn on the L4-L5 segment of the human waist, wherein the original motion data comprises three-axis acceleration data and three-axis angular velocity data; using a Kalman filtering algorithm to filter noise and calibrate the coordinate system of the original motion data, to obtain standard waist motion characteristic data. The present application realizes accurate and continuous recognition of the waist posture in daily scenarios, completes quantitative evaluation of the force on the lumbar vertebrae and prediction of the damage risk, has high detection accuracy and strong scene adaptability, and is portable and easy to use, which can provide a comprehensive basis for the user to manage the health of the waist and effectively prevent lumbar muscle damage.

Description

Technical Field

[0001] This invention relates to the field of waist posture recognition technology, and in particular to a method and system for human waist posture recognition and stress state assessment based on inertial sensors. Background Technology

[0002] Lumbar spine and lumbar muscle injuries are a common health problem in the current population. Prolonged poor posture, sitting posture, and improper exercise can cause continuous stress or sudden impact on the lumbar spine and muscles, leading to muscle strain, lumbar disc herniation, and other conditions. Therefore, real-time and accurate detection and assessment of human posture and lumbar spine and lumbar muscle stress are of great significance for preventing lumbar spine and lumbar muscle injuries.

[0003] In existing technologies, human posture detection is mainly divided into two categories: contact and non-contact. Non-contact detection (such as visual imaging technology) is greatly affected by factors such as ambient light and occlusion, making it difficult to achieve continuous monitoring in daily scenarios. In contact detection, traditional wearable devices can only achieve simple forward and backward posture judgment (such as whether the torso is leaning forward or backward), and lack quantitative assessment of the force on the lumbar spine and lumbar muscles.

[0004] In assessing lumbar spine and lumbar muscle stress, current biomechanical analyses largely rely on large-scale equipment in laboratory settings (such as motion capture systems and force sensor platforms), making portable, real-time assessment impossible. Furthermore, existing technologies do not employ differentiated mechanical models for different human states (resting and moving), resulting in insufficient accuracy in stress assessment and difficulty in predicting the impact of short-term impacts and long-term cumulative effects on the lumbar spine and lumbar muscles. In addition, existing solutions lack systematic analysis of daily posture data, failing to provide users with comprehensive reports on postural health, lumbar muscle fatigue, and lumbar spine injury risk.

[0005] Therefore, there is an urgent need for a technical solution that can continuously detect human posture in daily scenarios and combine differentiated biomechanical models to quantitatively assess the stress and cumulative effects on the lumbar spine and lumbar muscles, so as to make up for the shortcomings of existing technologies. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a method and system for human waist posture recognition and force state assessment based on inertial sensors.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for human waist posture recognition and force state assessment based on inertial sensors, comprising the following steps:

[0008] Raw motion data is collected in real time by an inertial measurement unit worn on the L4-L5 segment of the human waist. The raw motion data includes triaxial acceleration data and triaxial angular velocity data. The Kalman filter algorithm is used to filter out noise and calibrate the coordinate system of the raw motion data to obtain standard waist motion characteristic data.

[0009] Based on the acceleration amplitude change rate and angular velocity amplitude characteristics in the standard waist motion characteristic data, the current human motion state category is identified, and the human motion state category is divided into static maintenance state and dynamic activity state.

[0010] When the identification result is the static maintenance state, the current three-dimensional posture angle of the lumbar spine is calculated based on the fusion of the standard lumbar motion feature data. When the identification result is the dynamic activity state, the motion impact feature component and motion cycle parameter of the standard lumbar motion feature data in the vertical gravity direction are extracted.

[0011] Retrieve pre-stored user body shape parameters, construct a differentiated lumbar spine biomechanical model based on the user body shape parameters, input the three-dimensional posture angle of the lumbar spine or the motion impact characteristic component into the differentiated lumbar spine biomechanical model, and output the lumbar spine force load value.

[0012] The lumbar spine stress load values ​​over a continuous time period are integrated and analyzed to calculate the lumbar cumulative injury index. Based on the lumbar cumulative injury index, a posture health score and fatigue warning signal are generated, and the posture health score is displayed through an interactive terminal.

[0013] Preferably, the steps for using the Kalman filter algorithm to perform noise filtering and coordinate system calibration on the original motion data to obtain standard waist motion feature data are as follows:

[0014] The sampling frequency of the inertial measurement unit is set to a preset sampling frequency, and the inertial measurement unit collects raw triaxial acceleration data and raw triaxial angular velocity data according to the preset sampling frequency;

[0015] A Kalman filter is established, with attitude quaternions as the core state variables in its state equation. The raw triaxial angular velocity data is input into the state equation, and the angle change is obtained by integrating the raw triaxial angular velocity data with the sampling time. Attitude prediction is then completed by combining the quaternion multiplication operation. The observation equation of the Kalman filter uses the raw triaxial acceleration data as the observation vector. The pitch and roll angles are derived by back-calculating the gravity vector corresponding to the raw triaxial acceleration data, and the pitch and roll angles are used as observation values ​​to complete the measurement update. The Kalman filter is controlled to iteratively execute the above prediction update process and measurement update process, filtering out gyroscope drift noise and high-frequency oscillation noise in the raw triaxial angular velocity data, and outputting the optimal attitude quaternion.

[0016] Using a preset quaternion attitude calculation algorithm, based on the optimal attitude quaternion output by the Kalman filter, the raw triaxial acceleration data in the sensor coordinate system of the inertial measurement unit is transformed to the geodetic coordinate system. The direction of the gravitational acceleration vector is aligned and the gravitational component is subtracted to obtain the pure motion acceleration vector in the geodetic coordinate system. At the same time, the raw triaxial angular velocity data is corrected based on the optimal attitude quaternion to obtain a smoothed angular velocity vector. The output is standard waist motion feature data containing the pure motion acceleration vector in the geodetic coordinate system and the smoothed angular velocity vector.

[0017] Preferably, the step of identifying the current human motion state category based on the acceleration amplitude change rate and angular velocity amplitude characteristics in the standard waist motion characteristic data specifically includes:

[0018] Calculate the resultant acceleration amplitude and resultant angular velocity amplitude of the standard waist motion characteristic data within a preset time window;

[0019] Determine whether the magnitude of the resultant acceleration is within the gravitational acceleration threshold range, and simultaneously determine whether the magnitude of the resultant angular velocity is less than a preset rest angular velocity threshold;

[0020] If the magnitude of the resultant acceleration is within the gravitational acceleration threshold range and the magnitude of the resultant angular velocity is less than the stationary angular velocity threshold, then the human motion state is determined to be the static maintenance state.

[0021] If the rate of change of the resultant acceleration amplitude exceeds a preset motion rate of change threshold, and the resultant angular velocity amplitude is greater than the stationary angular velocity threshold, then the human motion state is determined to be the dynamic activity state.

[0022] Preferably, when the identification result is the static maintenance state, the specific steps for calculating the current three-dimensional posture angle of the lumbar spine based on the fusion of the standard lumbar motion feature data are as follows:

[0023] Extract the projection ratio of the gravity component on the three axes from the standard waist motion feature data;

[0024] The pitch angle of the lumbar spine relative to the vertical plane is calculated using inverse trigonometric functions, and the pitch angle is defined as the angle value of lumbar spine forward tilt or backward tilt.

[0025] The roll angle of the lumbar spine relative to the horizontal plane is calculated using inverse trigonometric functions, and the roll angle is defined as the angle value of the lumbar spine tilting to the left or right.

[0026] By integrating the angular velocity vector in the standard lumbar motion characteristic data over time, the heading angle of the lumbar spine relative to the initial orientation is calculated.

[0027] The pitch angle, roll angle, and yaw angle are combined to generate the three-dimensional attitude angles of the lumbar spine that describe the current spatial morphology of the lumbar spine.

[0028] Preferably, the steps of retrieving pre-stored user body shape parameters, constructing a differentiated lumbar spine biomechanical model based on the user body shape parameters, inputting the lumbar spine three-dimensional posture angle or the motion impact characteristic component into the differentiated lumbar spine biomechanical model, and outputting the lumbar spine force load value are as follows:

[0029] Obtain the user's body shape parameters, which include the user's height data and user's weight data;

[0030] When in the static maintenance state, the static mechanics calculation logic is invoked to estimate the upper body center of gravity lever arm length based on the user's height data, estimate the upper body mass based on the user's weight data, and construct the lumbar spine compression and shear force formulas that include time fatigue factors based on the principles of force balance and torque balance.

[0031] Substituting the three-dimensional posture angles of the lumbar spine into the static compression and shear force formula, the equivalent compressive force of the intervertebral disc, which serves as the load value on the lumbar spine, is calculated. and equivalent shear force ;

[0032] The static force correction formula is as follows:

[0033] ;

[0034] ;

[0035] in This represents the equivalent compressive force of the lumbar intervertebral disc, expressed in Newtons. This refers to the user's weight data. Represents the gravitational acceleration constant. This represents the preset upper body mass percentage coefficient. This indicates the length of the center of gravity lever arm calculated based on the user's height data. This represents the forward tilt angle in the three-dimensional posture angles of the lumbar spine. Indicates its absolute value. This indicates the preset effective lever arm distance for the lower back muscles. This represents the preset nonlinear growth coefficient of muscle fatigue. Indicates the duration of maintaining the current posture;

[0036] When in the dynamic activity state, the dynamic mechanics calculation logic is invoked, and the instantaneous impact force on the lumbar spine is calculated by combining the momentum theorem with the motion impact characteristic components, which is used as the force load value of the lumbar spine.

[0037] Preferably, when the identification result is the dynamic activity state, the specific steps for extracting the motion impact feature components and motion period parameters of the standard waist motion feature data in the vertical gravity direction are as follows:

[0038] Principal component analysis was performed on the standard waist motion characteristic data to separate the vertical acceleration waveform in the direction of gravity;

[0039] Identify the peaks and troughs in the vertical acceleration waveform, calculate the time difference between adjacent peaks, and obtain the motion period parameters.

[0040] The velocity change per unit time is obtained by integrating the bands in the vertical acceleration waveform that exceed the preset safety threshold.

[0041] By combining the user's weight data in the user's body shape parameters, the change in momentum corresponding to the change in velocity is calculated, and the change in momentum is determined as the motion impact characteristic component.

[0042] Preferably, the steps of performing integral analysis on the lumbar spine stress load values ​​over a continuous time period to calculate the lumbar cumulative injury index, and generating a postural health score and fatigue warning signal based on the lumbar cumulative injury index are as follows:

[0043] The cumulative duration of the lumbar spine stress load value within different preset pressure ranges within the continuous time period is statistically analyzed.

[0044] The cumulative duration of each preset pressure zone is multiplied by the corresponding weighting coefficient and summed to obtain the cumulative lumbar injury index.

[0045] Calculate the ratio of the cumulative lumbar injury index to a preset lumbar limit threshold. If the ratio exceeds a preset warning line, generate the fatigue warning signal.

[0046] The percentage of time the three-dimensional posture angles of the lumbar spine were within the normal range was statistically analyzed, and combined with the cumulative lumbar injury index, a weighted normalization calculation was performed to obtain the posture health score on a percentage scale.

[0047] Preferably, the step of displaying the posture health score through the interactive terminal specifically includes:

[0048] Establish a wireless data transmission link between the inertial measurement unit and the interactive terminal, and send the attitude health score and the fatigue warning signal to the interactive terminal;

[0049] The change curve of the lumbar spine stress load value is plotted on the display interface of the interactive terminal, and the posture health score is updated and displayed in real time in numerical form.

[0050] When the fatigue warning signal is received, the interactive terminal is controlled to issue a vibration prompt or pop up a text notification suggesting rest, thus completing the health intervention for the user.

[0051] The present invention also provides a system comprising:

[0052] The data acquisition module is used to control the inertial measurement unit worn on the L4-L5 segment of the human waist to acquire raw motion data in real time. The raw motion data includes triaxial acceleration data and triaxial angular velocity data.

[0053] The preprocessing and state recognition module is used to filter out noise and calibrate the coordinate system of the original motion data using the Kalman filter algorithm to obtain standard waist motion feature data, and to identify the current human motion state category based on the standard waist motion feature data.

[0054] The posture and parameter calculation module is used to calculate the three-dimensional posture angle of the lumbar spine when the human motion state category is a static maintenance state, and to extract the motion impact feature components when it is a dynamic activity state.

[0055] The biomechanical assessment module is used to retrieve pre-stored user body shape parameters, construct a differentiated lumbar spine biomechanical model, input the three-dimensional posture angle of the lumbar spine or the motion impact characteristic component into the differentiated lumbar spine biomechanical model, and calculate and output the force load value of the lumbar spine.

[0056] The health analysis and interaction module is used to calculate the cumulative injury index of the lower back, generate a postural health score and a fatigue warning signal, and display the postural health score through a visual interface.

[0057] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0058] (1) The present invention has high detection accuracy and strong scene adaptability: the IMU wearable device is fixed to the lumbar spine and directly collects motion data of the lumbar spine area. Combined with acceleration and angular velocity characteristics, it realizes accurate identification of state and posture. It is not affected by ambient light or occlusion and can realize continuous monitoring in daily scenes.

[0059] (2) The stress assessment of this invention is highly targeted and accurate: Differentiated biomechanical models (static / dynamic models) are used for static states (standing, sitting) and dynamic states. Combined with basic parameters such as human height and weight, the stress on the lumbar spine and lumbar muscles can be accurately quantified in different scenarios. In particular, the impact of impulse on the lumbar spine during exercise is considered, which makes up for the shortcomings of existing technologies that can only assess static stress. In terms of lumbar intervertebral disc stress assessment, pressure and shear force data can be given based on model algorithms to accurately capture the degree and direction of lumbar spine injury risk.

[0060] (3) The present invention realizes long-term cumulative effect assessment: by analyzing daily continuous data, it can not only provide real-time stress assessment, but also obtain comprehensive assessment results such as posture health and fatigue, predict the potential impact of long-term poor posture and stress on the lumbar spine and lumbar muscles, and provide users with a comprehensive health management basis.

[0061] (4) The present invention is highly portable and easy to use: it adopts a lightweight IMU wearable device, which is small in size, comfortable to wear, and convenient for data transmission. Data viewing and evaluation results can be obtained through ordinary terminal devices (mobile APP), making it suitable for daily promotion and use. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0064] Please see Figure 1 This invention provides a technical solution for human waist posture recognition and force state assessment based on inertial sensors, comprising the following steps:

[0065] Raw motion data is collected in real time by an inertial measurement unit worn on the L4-L5 segment of the human waist. The raw motion data includes triaxial acceleration data and triaxial angular velocity data. The Kalman filter algorithm is used to filter out noise and calibrate the coordinate system of the raw motion data to obtain standard waist motion characteristic data.

[0066] Based on the acceleration amplitude change rate and angular velocity amplitude characteristics in the standard waist motion characteristic data, the current human motion state category is identified, and the human motion state category is divided into static maintenance state and dynamic activity state.

[0067] When the identification result is a static state, the current three-dimensional posture angle of the lumbar spine is calculated based on the fusion of standard lumbar motion feature data. When the identification result is a dynamic state, the motion impact feature components and motion cycle parameters of the standard lumbar motion feature data in the vertical gravity direction are extracted.

[0068] Retrieve pre-stored user body shape parameters, construct a differentiated lumbar spine biomechanical model based on user body shape parameters, input the three-dimensional posture angle or motion impact characteristic components of the lumbar spine into the differentiated lumbar spine biomechanical model, and output the lumbar spine force load value.

[0069] The system performs integral analysis on the lumbar spine stress load values ​​over a continuous time period to calculate the lumbar cumulative injury index. Based on the lumbar cumulative injury index, it generates a posture health score and a fatigue warning signal, and displays the posture health score through an interactive terminal.

[0070] In this embodiment, the steps for using the Kalman filter algorithm to filter noise and calibrate the coordinate system of the raw motion data to obtain standard waist motion characteristic data are as follows: The sampling frequency of the inertial measurement unit (IMU) is set to a preset sampling frequency. The IMU collects raw triaxial acceleration and triaxial angular velocity data according to the preset sampling frequency. A Kalman filter is established, with attitude quaternions as the core state variables in the state equation. The raw triaxial angular velocity data is input into the state equation. The angle change is obtained by integrating the raw triaxial angular velocity data with the sampling time. Attitude prediction is completed by combining quaternion multiplication. The observation equation of the Kalman filter uses the raw triaxial acceleration data as the observation vector. The pitch and roll angles are obtained by back-calculating the gravity vector corresponding to the raw triaxial acceleration data. The pitch and roll angles are then... Using the angle as the observed value, the measurement update is completed. The Kalman filter is controlled to iteratively execute the above prediction update process and measurement update process, filtering out gyroscope drift noise in the raw three-axis angular velocity data and high-frequency oscillation noise in the raw three-axis acceleration data, and outputting the optimal attitude quaternion. Using a preset quaternion attitude calculation algorithm, based on the optimal attitude quaternion output by the Kalman filter, the raw three-axis acceleration data in the sensor coordinate system of the inertial measurement unit is transformed to the geodetic coordinate system. The direction of the gravitational acceleration vector is aligned and the gravity component is subtracted to obtain the pure motion acceleration vector in the geodetic coordinate system. At the same time, the raw three-axis angular velocity data is corrected based on the optimal attitude quaternion to obtain a smoothed angular velocity vector. The standard waist motion feature data containing the pure motion acceleration vector in the geodetic coordinate system and the smoothed angular velocity vector is output.

[0071] Specifically, the Kalman filter algorithm is used to filter noise and calibrate the coordinate system of the raw motion data. The sampling frequency of the inertial measurement unit is set to a preset sampling frequency. Combining the Nyquist sampling theorem and the waveform characteristics of daily human motion, in order to preserve complete high-frequency dynamic details and prevent frequency aliasing, the preset sampling frequency is set to 5 times the highest frequency of human motion, for example, a preset sampling frequency of 100 Hz. 100 times per second of raw triaxial acceleration and triaxial angular velocity data are uniformly collected. A Kalman filter is established, with attitude quaternions used as the core state variables of the Kalman filter. A first-order Markov state transition matrix is ​​constructed, and the raw triaxial angular velocity data are input into the state transition matrix. A fourth-order Runge library is used. The tower algorithm performs numerical integration on the raw triaxial angular velocity data and sampling time to obtain quaternion angle changes. Quaternion multiplication is then used to update the prior attitude estimate from the previous moment, completing attitude prediction. In this process, a process noise covariance matrix is ​​defined, where the diagonal elements represent the random walk noise variance of the gyroscope. This covariance is obtained by collecting one hour of continuous baseline data from the gyroscope in a stationary state and calculating its statistical variance; for example, a value of 0.0001 is set. The Kalman filter's observation equation uses the raw triaxial acceleration data as the observation vector, extracting the three upward components (X-axis, Y-axis, and Z-axis) of the raw triaxial acceleration data in the sensor's local coordinate system. Inverse trigonometric functions are then used to calculate the original triaxial acceleration. The tilt angle of the gravity vector corresponding to the initial data in three-dimensional space is used to deduce the pitch and roll angles. These angles are then fed into the observation model of the Kalman filter as observed values. Simultaneously, a measurement noise covariance matrix is ​​defined, with its diagonal elements representing the measurement noise variance of the accelerometer. This variance is also obtained through statistical analysis of the acceleration waveform variance under stationary conditions, for example, with a value set to 0.001. Based on the process noise covariance matrix and the measurement noise covariance matrix, the Kalman gain matrix is ​​calculated. This matrix is ​​used to evaluate the residual weights between the observed and predicted values. The residual components multiplied by the Kalman gain are then compensated to the prior attitude estimate, completing the measurement update. The Kalman filter is then controlled at a preset frequency of 100 Hz. The above prediction and measurement update processes are continuously iterated at the sampling frequency. In each iteration, the attitude error is continuously corrected, and the gyroscope drift noise accumulated over time in the raw three-axis angular velocity data and the high-frequency oscillation noise caused by human footsteps in the raw three-axis acceleration data are filtered out. The optimal attitude quaternion is output, and the four components of the optimal attitude quaternion are extracted to construct a direction cosine rotation matrix. The raw three-axis acceleration data in the sensor coordinate system of the inertial measurement unit is multiplied by this direction cosine rotation matrix and transformed to the geodetic coordinate system so that the direction of the acceleration vector is completely aligned with the real three-dimensional spatial coordinate system. In the direction perpendicular to the Z-axis of the geodetic coordinate system, the local gravitational acceleration constant 9 is subtracted.8. Align the direction of the gravitational acceleration vector and subtract the gravitational component to extract the acceleration component purely generated by the human torso motion. Simultaneously, correct the raw triaxial angular velocity data based on the optimal attitude quaternion. Utilize the gyroscope zero-bias state variable estimated during the Kalman filter iteration process, subtract this zero-bias state variable from the raw triaxial angular velocity data to obtain a smoothed angular velocity vector. Output standard waist motion feature data containing the pure motion acceleration vector in the geodetic coordinate system and the smoothed angular velocity vector.

[0072] In this embodiment, the steps for identifying the current human motion state category based on the acceleration amplitude change rate and angular velocity amplitude characteristics in the standard waist motion feature data are as follows: calculate the resultant acceleration amplitude and resultant angular velocity amplitude of the standard waist motion feature data within a preset time window, determine whether the resultant acceleration amplitude is within the gravitational acceleration threshold range, and simultaneously determine whether the resultant angular velocity amplitude is less than a preset rest angular velocity threshold. If the resultant acceleration amplitude is within the gravitational acceleration threshold range and the resultant angular velocity amplitude is less than the rest angular velocity threshold, then the human motion state category is determined to be a stationary state. If the rate of change of the resultant acceleration amplitude exceeds a preset motion change rate threshold and the resultant angular velocity amplitude is greater than the rest angular velocity threshold, then the human motion state category is determined to be a dynamic activity state.

[0073] Specifically, the resultant acceleration amplitude and resultant angular velocity amplitude of the standard waist motion characteristic data are calculated within a preset time window. This preset time window is a sliding window ranging from 0.5 seconds to 1.0 seconds, encompassing... Each sampling point, sum of acceleration amplitude The calculation is the mean of the acceleration magnitudes of all sampling points within the window, i.e. ,in This represents the average magnitude of the calculated resultant acceleration. This indicates the total number of data points collected within the preset time window. Indicates the number of elements in the current window. The index number of each sampling point Indicates the first The acceleration component values ​​of each sampling point along the X-axis of the sensor. Indicates the first The acceleration component values ​​of each sampling point along the Y-axis of the sensor. Indicates the first The acceleration component values ​​and resultant angular velocity amplitudes at each sampling point along the Z-axis of the sensor. Similarly, the calculation determines whether the resultant acceleration amplitude is within the gravitational acceleration threshold range, which is set as follows: ,in This refers to the local gravitational acceleration, for example, 9.8. , To allow for acceptable sensor noise and minor vibration tolerance, data is collected after the sensor has been stationary for 10 minutes, and its standard deviation is calculated to be three times the standard deviation. For example, a value of 0.2 is set. Simultaneously, it determines whether the magnitude of the resultant angular velocity is less than a preset stationary angular velocity threshold, which is set to twice the sensor's zero-bias stability index, for example, 0.05. If the resultant acceleration amplitude is within the gravitational acceleration threshold range and the resultant angular velocity amplitude is less than the rest angular velocity threshold, then the human motion state is determined to be a stationary state. If the rate of change of the resultant acceleration amplitude exceeds a preset motion rate of change threshold, this rate of change is determined by calculating the variance of the acceleration magnitude within the window. To characterize this, the threshold for the rate of change of motion is set by analyzing the acceleration fluctuation characteristics during human walking. For example, the lower quartile of the statistical distribution of acceleration variance during walking can be taken and set to 0.5. If the resultant angular velocity amplitude is greater than the static angular velocity threshold, then the human motion state is determined to be a dynamic activity state.

[0074] In this embodiment, when the identification result is a static state, the specific steps for calculating the current three-dimensional posture angle of the lumbar spine based on the fusion of standard lumbar motion feature data are as follows: extract the projection ratio of the gravity component on the three axes in the standard lumbar motion feature data, calculate the pitch angle of the lumbar spine relative to the vertical plane using inverse trigonometric functions, define the pitch angle as the angle value of the lumbar spine tilting forward or backward, calculate the roll angle of the lumbar spine relative to the horizontal plane using inverse trigonometric functions, define the roll angle as the angle value of the lumbar spine tilting to the left or right, perform time integration using the angular velocity vector in the standard lumbar motion feature data, calculate the heading angle of the lumbar spine relative to the initial orientation, and combine the pitch angle, roll angle and heading angle to generate the three-dimensional posture angle of the lumbar spine describing the current spatial morphology of the lumbar spine.

[0075] Specifically, the projection ratios of the gravity component in the standard waist motion feature data onto the three axes are extracted, that is, the components of the calibrated acceleration vector on the three axes of the sensor coordinate system are obtained. Since the accelerometer primarily measures the gravitational component when stationary, these components are used to calculate the tilt angle. The inverse trigonometric functions are then used to calculate the pitch angle of the lumbar spine relative to the vertical plane. The formula is as follows: ,in This represents the calculated pitch angle. The gravitational acceleration components corresponding to the anterior-posterior direction of the spine. The corresponding lateral upward gravitational acceleration components of the spine. Corresponding to the vertical component of gravitational acceleration, the pitch angle is defined as the angle of lumbar lordosis (forward or backward tilt). The roll angle of the lumbar spine relative to the horizontal plane is calculated using inverse trigonometric functions. The formula is as follows: ,in This represents the calculated roll angle. and The meaning is the same as above. The roll angle is defined as the angle of lateral or left lateral tilt of the lumbar spine. The angular velocity vector in the standard lumbar motion characteristic data is used for time integration. The trapezoidal integration method or Runge-Kutta method is used to measure the calibrated Z-axis angular velocity. To perform the integration, the initial heading angle is set to 0 degrees or initialized using magnetometer data. The integration formula is as follows: ,in Indicates the current The heading angle at any moment, Indicates the previous moment The heading angle, Indicates the current The angular velocity value of the sensor along the Z-axis at any given time. The time interval between two samplings is represented by the complementary filtering algorithm combined with magnetometer data (if available) to correct the integral drift, calculate the heading angle of the lumbar spine relative to the initial orientation, and combine the pitch angle, roll angle and heading angle to generate the three-dimensional attitude angle of the lumbar spine describing the current spatial morphology of the lumbar spine.

[0076] In this embodiment, the steps of retrieving pre-stored user body shape parameters, constructing a differentiated lumbar spine biomechanical model based on these parameters, and inputting the lumbar spine three-dimensional posture angle or motion impact characteristic components into the differentiated lumbar spine biomechanical model to output the lumbar spine force load value are as follows: The user body shape parameters, including user height and weight data, are obtained; when in the static maintenance state, static mechanics calculation logic is invoked, and the upper body center of gravity lever arm length is estimated based on the user height data, and the upper body mass is estimated based on the user weight data; a lumbar spine compression and shear force formula, including a time fatigue factor, is constructed based on the principles of force balance and torque balance; the lumbar spine three-dimensional posture angle is substituted into the static compression and shear force formula to calculate the lumbar intervertebral disc equivalent compressive force, which serves as the lumbar spine force load value. and equivalent shear force The static force correction formula is as follows: ; ;in This represents the equivalent compressive force of the lumbar intervertebral disc, expressed in Newtons. This refers to the user's weight data. Represents the gravitational acceleration constant. This represents the preset upper body mass percentage coefficient. This indicates the length of the center of gravity lever arm calculated based on the user's height data. This represents the forward tilt angle in the three-dimensional posture angles of the lumbar spine. Indicates its absolute value. This indicates the preset effective lever arm distance for the lower back muscles. This represents the preset nonlinear growth coefficient of muscle fatigue. This indicates the duration of maintaining the current posture; when in the dynamic activity state, the dynamic mechanics calculation logic is invoked, and the instantaneous impact force on the lumbar spine is calculated using the momentum theorem combined with the motion impact characteristic components, which is used as the force load value of the lumbar spine.

[0077] Specifically, the system retrieves pre-stored user body shape parameters, analyzes the included user height and weight data, and, while maintaining a static state, invokes static mechanics calculation logic. It extracts the user's height data and multiplies it by a preset center-of-gravity height ratio coefficient to estimate the upper body's center-of-gravity lever arm length. This ratio coefficient, based on anthropometry statistics, is 0.35; for a height of 1.8 meters, the lever arm length is 0.63 meters. It then extracts weight data to estimate the upper body mass. Combining the principles of force and torque balance, it introduces a time decay variable to construct a static force formula. Substituting the previously obtained lumbar spine three-dimensional posture angles into the formula, it calculates the equivalent compressive force on the lumbar intervertebral disc. and equivalent shear force The formula is as well as ,in This represents the equivalent compressive force of the lumbar intervertebral disc, measured in Newtons. It is a quantitative physical index reflecting the degree of vertical compression experienced by the spine in the axial direction, and is obtained by adding the mechanical components in the formula. It represents the equivalent shear force of the lumbar intervertebral disc, measured in Newtons. It reflects the slippage tendency force in the horizontal cross-section of the spine and is a key force dimension indicator for assessing induced skeletal strain injuries. This represents the user's weight data, in kilograms. It is read from the terminal database and is the user's actual physical weight value pre-filled through the interface during initial use. This directly determines the static load of the foundation. This represents the gravitational acceleration constant, which indicates the downward vertical acceleration exerted by Earth's gravity on the mass of various parts of the human body. Under typical altitude living and working conditions, it is fixed at a value of 9.8. This represents the preset upper body mass percentage coefficient, calculated based on the statistical average of a human body shape distribution database and set to 0.65. It refers to the proportion of the upper body weight to the total body weight. This refers to the length of the center of gravity lever arm calculated based on the user's height data, in meters. It represents the vertical physical distance from the biomechanical rotation center of the L4 and L5 lumbar vertebrae to the center of gravity of the upper body. This represents the forward tilt angle in the three-dimensional posture angles of the lumbar spine, reflecting the degree of forward bending or backward tilting of the human torso relative to the vertical plane of a plumb bob. It is output from the matrix calculation mentioned earlier. The result of the calculation representing the absolute value of the forward tilt angle is mainly to eliminate the difference in sign calculation caused by the forward and backward pitch directions. Negative backward tilt angles are uniformly converted to positive values ​​before being substituted into the trigonometric functions. This represents the preset effective lever arm distance for the lumbar and back muscle groups, used to equivalently replace the muscle tension lever arm generated by the erector spinae muscles when counteracting gravity. Based on the anatomical sample from MRI images, the mean value is set to 0.06 meters. This represents the preset nonlinear growth coefficient of muscle fatigue, used to simulate the degree of increased lumbar spine stress compensation due to muscle fatigue over time. It was set to 0.15 based on statistical analysis of the decrease in the median frequency of electromyographic signals. This indicates the duration of maintaining the current static posture, in seconds. It is updated in real time by continuously accumulating the physical duration of the sampling period when the human body is in a static state through a built-in timer. When in a dynamic activity state, the dynamic mechanics calculation logic is invoked to extract the motion impact feature component containing the momentum change. This momentum change is divided by a preset foot contact buffer time parameter, which is set to 0.05 seconds by analyzing the test running and jumping video frame by frame. The momentum change rate is calculated and converted into instantaneous vertical impact force, which is used as the lumbar spine force load value in this state.

[0078] In this embodiment, when the identification result is a dynamic activity state, the specific steps for extracting the motion impact feature components and motion period parameters of the standard waist motion feature data in the vertical gravity direction are as follows: perform principal component analysis on the standard waist motion feature data, separate the vertical acceleration waveform in the gravity direction, identify the peaks and troughs in the vertical acceleration waveform, calculate the time difference between adjacent peaks to obtain the motion period parameters, perform integral calculation on the bands in the vertical acceleration waveform that exceed the preset safety threshold to obtain the velocity change per unit time, combine the user's weight data in the user's body shape parameters to calculate the momentum change corresponding to the velocity change, and determine the momentum change as the motion impact feature component.

[0079] Specifically, principal component analysis is performed on standard lumbar motion characteristic data to construct a covariance matrix of acceleration data. Eigenvalue decomposition is then performed on the covariance matrix, and the eigenvector corresponding to the largest eigenvalue is selected as the principal axis of human motion. This direction is approximately parallel to the direction of gravity in upright motion. The triaxial acceleration is projected onto this principal axis to separate the vertical acceleration waveform in the direction of gravity. Peaks and troughs in the vertical acceleration waveform are identified. A sliding window extreme value detection method is used, and a minimum peak spacing threshold is set to avoid multi-peak interference, for example, 0.3 seconds. The time difference between adjacent peaks is calculated, and the average time difference of multiple consecutive gait cycles is statistically analyzed to obtain the motion cycle parameter. Integral calculations are performed on the segments of the vertical acceleration waveform that exceed a preset safety threshold, set to 1.5g (g is gravitational acceleration). This threshold is used to filter out the landing moments or jump landing phases that have a significant impact on the lumbar spine. Numerical integration is performed on the acceleration-time curve portion exceeding this threshold to obtain the velocity change per unit time, i.e., the impact velocity. Combining user weight data from the user's body shape parameters, using the formula Calculate the impulse, where This represents the calculated impulse value. To ensure effective impact mass, a certain percentage (e.g., 80%) of the user's body weight is included in the impact calculation. The velocity change obtained by integration is used to calculate the momentum change corresponding to the velocity change, and the momentum change is determined as the characteristic component of the motion impact.

[0080] In this embodiment, the steps of performing integral analysis on the lumbar spine stress load values ​​over a continuous time period to calculate the lumbar cumulative injury index and generating a posture health score and fatigue warning signal based on the lumbar cumulative injury index are as follows: The cumulative duration of the lumbar spine stress load values ​​within different preset pressure ranges over a continuous time period is statistically analyzed. The cumulative duration of each preset pressure range is multiplied by its corresponding weight coefficient and summed to obtain the lumbar cumulative injury index. The ratio of the lumbar cumulative injury index to a preset muscle endurance limit threshold is calculated. If the ratio exceeds a preset warning line, a fatigue warning signal is generated. The percentage of time the lumbar spine three-dimensional posture angle is within the normal range is statistically analyzed. Combined with the lumbar cumulative injury index, a percentage posture health score is obtained through weighted normalization.

[0081] Specifically, the cumulative duration of lumbar spine stress load values ​​within different preset pressure ranges over a continuous time period is statistically analyzed. These pressure ranges are divided into low-load (<1000N), medium-load (1000N-3000N), and high-load (>3000N) ranges. The number of sample points falling within each range is recorded and converted into duration. The cumulative duration of each preset pressure range is multiplied by its corresponding weighting coefficient and summed. The weighting coefficient is non-linearly set based on the risk of intervertebral disc damage under different loads; for example, the low-load coefficient... Medium load factor High load factor The calculation formula is: ,in Indicates the cumulative injury index of the lower back. This indicates the total number of pressure ranges. This indicates that the force load value falls within the first... The cumulative duration within each pressure range Indicates the first The risk weight coefficients corresponding to each pressure range are used to obtain the lumbar cumulative injury index. The ratio of the lumbar cumulative injury index to the preset muscle endurance limit threshold is calculated. The muscle endurance limit threshold is calculated for individual parameters based on the Rohmert curve or related fatigue model, and represents the total load that reaches the point of exhaustion or injury risk. If the ratio exceeds the preset warning line, for example, if the ratio is greater than 0.8 (i.e., reaching 80% of the endurance limit), a fatigue warning signal is generated. The percentage of time that the three-dimensional posture angles of the lumbar spine are within the normal range is counted. The normal range is defined as forward tilt angle <10 degrees and lateral tilt angle <5 degrees. Combined with the lumbar cumulative injury index, a weighted deduction system is used. The initial score is 100 points. For each time the injury index threshold is exceeded, the corresponding score is deducted. For every additional 10 minutes of substandard posture, the corresponding score is deducted. The posture health score is obtained on a percentage scale through weighted normalization.

[0082] In this embodiment, the steps of displaying the posture health score through the interactive terminal are as follows: establishing a wireless data transmission link between the inertial measurement unit and the interactive terminal, sending the posture health score and fatigue warning signal to the interactive terminal, plotting the change curve of the lumbar spine force load value on the display interface of the interactive terminal, and updating and displaying the posture health score in real time in numerical form. When a fatigue warning signal is received, controlling the interactive terminal to issue a vibration prompt or pop up a text notification suggesting rest, thus completing the health intervention for the user.

[0083] Specifically, a wireless data transmission link is established between the inertial measurement unit and the interactive terminal. Bluetooth Low Energy (BLE) technology is used, and a private GATT service is defined. The calculation results are encapsulated into hexadecimal data packets, and the posture health score and fatigue warning signal are sent to the interactive terminal. The change curve of the lumbar spine force load value is plotted on the interactive terminal's display interface. A fixed-length first-in-first-out (FIFO) data buffer queue is established, and the received real-time load values ​​are pushed into the queue. The canvas is refreshed in real time using a plotting library (such as MPAndroidChart or Core Plot) to display the force fluctuation in the past minute in the form of a line graph, and the posture health score is updated and displayed in real time in numerical form. When a fatigue warning signal is received, the status bit in the data packet is parsed. If the fatigue flag bit is 1, the vibration motor drive interface of the terminal is called to control the interactive terminal to issue a vibration prompt or pop up a text notification suggesting rest, thus completing the health intervention for the user.

[0084] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for human waist posture recognition and force state assessment based on inertial sensors, characterized in that, Includes the following steps: Raw motion data is collected in real time by an inertial measurement unit worn on the L4-L5 segment of the human waist. The raw motion data includes triaxial acceleration data and triaxial angular velocity data. The Kalman filter algorithm is used to filter out noise and calibrate the coordinate system of the raw motion data to obtain standard waist motion characteristic data. Based on the acceleration amplitude change rate and angular velocity amplitude characteristics in the standard waist motion characteristic data, the current human motion state category is identified, and the human motion state category is divided into static maintenance state and dynamic activity state. When the identification result is the static maintenance state, the current three-dimensional posture angle of the lumbar spine is calculated based on the fusion of the standard lumbar motion feature data. When the identification result is the dynamic activity state, the motion impact feature component and motion cycle parameter of the standard lumbar motion feature data in the vertical gravity direction are extracted. Retrieve pre-stored user body shape parameters, construct a differentiated lumbar spine biomechanical model based on the user body shape parameters, input the three-dimensional posture angle of the lumbar spine or the motion impact characteristic component into the differentiated lumbar spine biomechanical model, and output the lumbar spine force load value. The lumbar spine stress load values ​​over a continuous time period are integrated and analyzed to calculate the lumbar cumulative injury index. Based on the lumbar cumulative injury index, a posture health score and fatigue warning signal are generated, and the posture health score is displayed through an interactive terminal.

2. The method for human waist posture recognition and force state assessment based on inertial sensors according to claim 1, characterized in that, The specific steps for using the Kalman filter algorithm to filter noise and calibrate the coordinate system of the original motion data to obtain standard waist motion feature data are as follows: The sampling frequency of the inertial measurement unit is set to a preset sampling frequency, and the inertial measurement unit collects raw triaxial acceleration data and raw triaxial angular velocity data according to the preset sampling frequency; A Kalman filter is established, with attitude quaternions as the core state variables in its state equation. The raw triaxial angular velocity data is input into the state equation, and the angle change is obtained by integrating the raw triaxial angular velocity data with the sampling time. Attitude prediction is then completed by combining the quaternion multiplication operation. The observation equation of the Kalman filter uses the raw triaxial acceleration data as the observation vector. The pitch and roll angles are derived by back-calculating the gravity vector corresponding to the raw triaxial acceleration data, and the pitch and roll angles are used as observation values ​​to complete the measurement update. The Kalman filter is controlled to iteratively execute the above prediction update process and measurement update process, filtering out gyroscope drift noise and high-frequency oscillation noise in the raw triaxial angular velocity data, and outputting the optimal attitude quaternion. Using a preset quaternion attitude calculation algorithm, based on the optimal attitude quaternion output by the Kalman filter, the raw triaxial acceleration data in the sensor coordinate system of the inertial measurement unit is transformed to the geodetic coordinate system. The direction of the gravitational acceleration vector is aligned and the gravitational component is subtracted to obtain the pure motion acceleration vector in the geodetic coordinate system. At the same time, the raw triaxial angular velocity data is corrected based on the optimal attitude quaternion to obtain a smoothed angular velocity vector. The output is standard waist motion feature data containing the pure motion acceleration vector in the geodetic coordinate system and the smoothed angular velocity vector.

3. The method for human waist posture recognition and force state assessment based on inertial sensors according to claim 1, characterized in that, Based on the acceleration amplitude change rate and angular velocity amplitude characteristics in the standard waist motion feature data, the specific steps for identifying the current human motion state category are as follows: Calculate the resultant acceleration amplitude and resultant angular velocity amplitude of the standard waist motion characteristic data within a preset time window; Determine whether the magnitude of the resultant acceleration is within the gravitational acceleration threshold range, and simultaneously determine whether the magnitude of the resultant angular velocity is less than a preset rest angular velocity threshold; If the magnitude of the resultant acceleration is within the gravitational acceleration threshold range and the magnitude of the resultant angular velocity is less than the stationary angular velocity threshold, then the human motion state is determined to be the static maintenance state. If the rate of change of the resultant acceleration amplitude exceeds a preset motion rate of change threshold, and the resultant angular velocity amplitude is greater than the stationary angular velocity threshold, then the human motion state is determined to be the dynamic activity state.

4. The method for human waist posture recognition and force state assessment based on inertial sensors according to claim 1, characterized in that, When the identification result is the static maintenance state, the specific steps for calculating the current three-dimensional posture angle of the lumbar spine based on the fusion of the standard lumbar motion feature data are as follows: Extract the projection ratio of the gravity component on the three axes from the standard waist motion feature data; The pitch angle of the lumbar spine relative to the vertical plane is calculated using inverse trigonometric functions, and the pitch angle is defined as the angle value of lumbar spine forward tilt or backward tilt. The roll angle of the lumbar spine relative to the horizontal plane is calculated using inverse trigonometric functions, and the roll angle is defined as the angle value of the lumbar spine tilting to the left or right. By integrating the angular velocity vector in the standard lumbar motion characteristic data over time, the heading angle of the lumbar spine relative to the initial orientation is calculated. The pitch angle, roll angle, and yaw angle are combined to generate the three-dimensional attitude angles of the lumbar spine that describe the current spatial morphology of the lumbar spine.

5. The method for human waist posture recognition and force state assessment based on inertial sensors according to claim 1, characterized in that, The specific steps for retrieving pre-stored user body shape parameters, constructing a differentiated lumbar spine biomechanical model based on these parameters, inputting the lumbar spine three-dimensional posture angles or motion impact characteristic components into the differentiated lumbar spine biomechanical model, and outputting the lumbar spine force load values ​​are as follows: Obtain the user's body shape parameters, which include the user's height data and user's weight data; When in the static maintenance state, the static mechanics calculation logic is invoked to estimate the upper body center of gravity lever arm length based on the user's height data, estimate the upper body mass based on the user's weight data, and construct the lumbar spine compression and shear force formulas that include time fatigue factors based on the principles of force balance and torque balance. Substituting the three-dimensional posture angles of the lumbar spine into the static compression and shear force formula, the equivalent compressive force of the intervertebral disc, which serves as the load value on the lumbar spine, is calculated. and equivalent shear force ; The static force correction formula is as follows: ； ； in This represents the equivalent compressive force of the lumbar intervertebral disc, expressed in Newtons. This refers to the user's weight data. Represents the gravitational acceleration constant. This represents the preset upper body mass percentage coefficient. This indicates the length of the center of gravity lever arm calculated based on the user's height data. This represents the forward tilt angle in the three-dimensional posture angles of the lumbar spine. Indicates its absolute value. This indicates the preset effective lever arm distance for the lower back muscles. This represents the preset nonlinear growth coefficient of muscle fatigue. Indicates the duration of maintaining the current posture; When in the dynamic activity state, the dynamic mechanics calculation logic is invoked, and the instantaneous impact force on the lumbar spine is calculated by combining the momentum theorem with the motion impact characteristic components, which is used as the force load value of the lumbar spine.

6. The method for human waist posture recognition and force state assessment based on inertial sensors according to claim 1, characterized in that, When the identification result is the dynamic activity state, the specific steps for extracting the motion impact feature components and motion period parameters of the standard waist motion feature data in the vertical gravity direction are as follows: Principal component analysis was performed on the standard waist motion characteristic data to separate the vertical acceleration waveform in the direction of gravity; Identify the peaks and troughs in the vertical acceleration waveform, calculate the time difference between adjacent peaks, and obtain the motion period parameters. The velocity change per unit time is obtained by integrating the bands in the vertical acceleration waveform that exceed the preset safety threshold. By combining the user's weight data in the user's body shape parameters, the change in momentum corresponding to the change in velocity is calculated, and the change in momentum is determined as the motion impact characteristic component.

7. The method for human waist posture recognition and force state assessment based on inertial sensors according to claim 1, characterized in that, The specific steps for performing integral analysis on the lumbar spine stress load values ​​over a continuous time period, calculating the lumbar cumulative injury index, and generating a postural health score and fatigue warning signal based on the lumbar cumulative injury index are as follows: The cumulative duration of the lumbar spine stress load value within different preset pressure ranges within the continuous time period is statistically analyzed. The cumulative duration of each preset pressure zone is multiplied by the corresponding weighting coefficient and summed to obtain the cumulative lumbar injury index. Calculate the ratio of the cumulative lumbar injury index to a preset lumbar limit threshold. If the ratio exceeds a preset warning line, generate the fatigue warning signal. The percentage of time the three-dimensional posture angles of the lumbar spine were within the normal range was statistically analyzed, and combined with the cumulative lumbar injury index, a weighted normalization calculation was performed to obtain the posture health score on a percentage scale.

8. The method for human waist posture recognition and force state assessment based on inertial sensors according to claim 1, characterized in that, The specific steps for displaying the posture health score via the interactive terminal are as follows: Establish a wireless data transmission link between the inertial measurement unit and the interactive terminal, and send the attitude health score and the fatigue warning signal to the interactive terminal; The change curve of the lumbar spine stress load value is plotted on the display interface of the interactive terminal, and the posture health score is updated and displayed in real time in numerical form. When the fatigue warning signal is received, the interactive terminal is controlled to issue a vibration prompt or pop up a text notification suggesting rest, thus completing the health intervention for the user.

9. The system for human waist posture recognition and force state assessment based on inertial sensors according to any one of claims 1-8, characterized in that, include: The data acquisition module is used to control the inertial measurement unit worn on the L4-L5 segment of the human waist to acquire raw motion data in real time. The raw motion data includes triaxial acceleration data and triaxial angular velocity data. The preprocessing and state recognition module is used to filter out noise and calibrate the coordinate system of the original motion data using the Kalman filter algorithm to obtain standard waist motion feature data, and to identify the current human motion state category based on the standard waist motion feature data. The posture and parameter calculation module is used to calculate the three-dimensional posture angle of the lumbar spine when the human motion state category is a static maintenance state, and to extract the motion impact feature components when it is a dynamic activity state. The biomechanical assessment module is used to retrieve pre-stored user body shape parameters, construct a differentiated lumbar spine biomechanical model, input the three-dimensional posture angle of the lumbar spine or the motion impact characteristic component into the differentiated lumbar spine biomechanical model, and calculate and output the force load value of the lumbar spine. The health analysis and interaction module is used to calculate the cumulative injury index of the lower back, generate a postural health score and a fatigue warning signal, and display the postural health score through a visual interface.

Images