A method and system for realizing posture detection and early warning based on multimodal data

Abstract

The present invention relates to the field of posture detection technology, and specifically discloses a method and system for implementing posture detection and early warning based on multimodal data. The method comprises: acquiring multimodal data, wherein the multimodal data includes ambient air pressure data, user acceleration data, angular velocity data, height data, magnetic field data, and weight data; preprocessing the multimodal data to obtain preprocessed multimodal data; performing multi-stage posture detection based on the preprocessed multimodal data, and performing multimodal early warning based on the preprocessed multimodal data during the multi-stage posture detection process; wherein the multi-stage posture detection includes weightlessness detection, impact detection, and posture stability detection performed sequentially; and outputting the multi-stage posture detection results and multimodal early warning information. The present invention can significantly reduce false alarm rates and missed alarm rates, provide more personalized, timely, and accurate fall risk management, and thus greatly improve user safety and quality of life.

Description

Technical Field

[0001] The present invention relates to the technical field of posture detection, and in particular to a method for realizing posture detection and early warning based on multi-mode data and a system for realizing posture detection and early warning based on multi-mode data. Background Art

[0002] With the aging population and people's increasing emphasis on health and safety, the demand for high-precision, low-false-alarm fall detection technology is becoming increasingly urgent. Traditional fall detection systems, whether based on sensors or vision, generally have the following shortcomings:

[0003] Limited detection accuracy, high false positive and false negative rates: Most systems use fixed acceleration or posture thresholds, which cannot adapt to the physiological differences of different individuals (such as height and weight). This may result in missed detections for lighter users and false positives for heavier users or daily activities (such as sitting, jumping, and dropping the device).

[0004] Lack of personalized adaptability: Failure to fully consider the impact of user physiological parameters on fall characteristics results in the lack of universality of the detection model and the need for complex individual calibration.

[0005] Strong dependence on device wearing orientation: Inertial sensor data is greatly affected by the device wearing position and orientation. If it cannot be effectively converted to a unified reference frame, the robustness of the detection will be greatly reduced.

[0006] Only able to trigger an alarm after the event, lacking advance warning capabilities: Most existing systems can only trigger an alarm after a fall occurs, and are unable to identify potential risks before a fall occurs, thus missing the best time for prevention and intervention.

[0007] Limitations of a single criterion: Relying solely on a single sensor data or simple logical combination makes it difficult to fully capture the complex characteristics of a fall, and the discrimination is low.

[0008] Some research and products have attempted to introduce multi-sensor data for optimization, but these solutions still have the following technical drawbacks:

[0009] Insufficient data fusion depth: Sensor data are often processed independently or simply superimposed, lacking deep feature fusion and multi-dimensional cross-validation mechanisms.

[0010] The algorithm is not robust enough: it has weak anti-interference capabilities and is still prone to false alarms under complex daily activities and environmental interference.

[0011] Imperfect early warning mechanism: There is a lack of early risk prediction models based on multimodal data fusion, and hierarchical early warning and personalized risk management cannot be achieved.

[0012] Given these challenges, current fall detection technology faces an urgent need for technological upgrades. This requires a deep integration of multimodal sensing, adaptive algorithms, physical models, and machine learning to create a truly personalized, highly accurate, and robust intelligent fall detection system capable of providing early warning. This invention, proposed against this backdrop, overcomes several technical bottlenecks of traditional approaches and offers new solutions and development directions for the industry. Summary of the Invention

[0013] In response to the defects and shortcomings in the existing technology, the present invention provides a method for realizing posture detection and early warning based on multimodal data to solve the technical problems existing in the existing technology such as insufficient data fusion depth, poor algorithm robustness, and imperfect early warning mechanism.

[0014] As a first aspect of the present invention, a method for implementing posture detection and early warning based on multimodal data is provided, wherein the method for implementing posture detection and early warning based on multimodal data comprises:

[0015] Step S1: Acquire multimodal data, wherein the multimodal data includes ambient air pressure data, user acceleration data, user angular velocity data, user magnetic field data, user height data, and user weight data;

[0016] Step S2: preprocessing the multimodal data to obtain preprocessed multimodal data;

[0017] Step S3: performing multi-stage posture detection based on the pre-processed multi-mode data, and performing multi-modal early warning based on the pre-processed multi-mode data during the multi-stage posture detection process; wherein the multi-stage posture detection includes weightlessness detection, impact detection, and posture stability detection performed in sequence;

[0018] Step S4: Output multi-stage posture detection results and multi-modal warning information.

[0019] Furthermore, the step S1 further includes:

[0020] Acquire the user's acceleration data in real time through a three-axis accelerometer;

[0021] The user's angular velocity data is obtained in real time through the three-axis gyroscope;

[0022] Acquire the user's magnetic field data in real time through a three-axis magnetometer;

[0023] Obtain ambient air pressure data in real time through the barometer.

[0024] Furthermore, the step S2 further includes:

[0025] Step S21: Clean, normalize, and perform Kalman filtering on the acceleration data and angular velocity data of the user; wherein the state equation of the Kalman filter is expressed as:

[0026] ;

[0027] ;

[0028] in, For the moment t The system state vector of For the moment The system state vector of F is the state transfer matrix; B is the control input matrix; For the moment t The control vector of is the process noise; For the moment t The measurement vector, namely the user's acceleration data and the user's angular velocity data; is the measurement matrix; To measure noise;

[0029] Step S22: Constructing a three-dimensional world coordinate system by combining the gravity vector and the magnetic vector, and converting the filtered acceleration data and the filtered angular velocity data into the three-dimensional world coordinate system; wherein the construction process of the three-dimensional world coordinate system is as follows:

[0030] (1) Gravity vector axis determination: normalize the average value of the filtered acceleration data and define it as the Z axis in the three-dimensional world coordinate system;

[0031] ;

[0032] in, is the Z-axis unit vector in the three-dimensional world coordinate system; is the average value of the filtered gravity acceleration vector over a period of time; is the modulus of the average value of the gravitational acceleration vector, which is used to normalize the average value of the gravitational acceleration vector to a unit vector;

[0033] (2) Magnetic vector axis determination: Utilize the user's magnetic field data and the Z axis in the three-dimensional world coordinate system to determine the X axis in the three-dimensional world coordinate system, ensuring that the X axis in the three-dimensional world coordinate system is orthogonal to the Z axis;

[0034] ;

[0035] ;

[0036] in, is the three-dimensional world coordinate system axis unit vector; is the magnetic field vector measured by the three-axis magnetometer; is the magnetic field vector The Z-axis unit vector in the three-dimensional world coordinate system projection on; is a vector and vector The dot product of is a vector The square of the module length.

[0037] Furthermore, the multi-stage posture detection according to the pre-processed multimodal data further includes:

[0038] First, the baseline impact acceleration threshold is modified according to the user's weight data to obtain a modified impact acceleration threshold; and the baseline pitch angle threshold is modified according to the user's height data to obtain a modified pitch angle threshold; and the minimum height difference threshold is calculated according to the user's height data; wherein,

[0039] (1) The modified impact acceleration threshold The calculation formula is as follows:

[0040] ;

[0041] Where, is the baseline impact acceleration threshold, W is the user's weight in kg; 70 is the baseline weight in kg; 0.3 is the adjustment coefficient;

[0042] (2) The corrected pitch angle threshold The calculation formula is as follows:

[0043] ;

[0044] Where 75° is the reference pitch angle threshold, H is the user's height data, in cm; 170 is the base height, in cm; -0.2 is the adjustment coefficient;

[0045] (3) The minimum height difference threshold The calculation formula is as follows:

[0046] ;

[0047] Where, H The user's height data, in cm;

[0048] Second, firstly, weightlessness detection is performed based on the user's combined acceleration data; after the weightlessness detection is completed, impact detection is performed based on the user's combined acceleration data peak value and angular velocity data peak value; after the impact detection is completed, posture stability detection is performed based on the user's height change, pitch angle change, and roll angle change; wherein,

[0049] (1) Weightlessness detection: Determine whether the user's combined acceleration data is less than the preset weightlessness threshold for a period of time. If so, it means that the user is currently in a weightlessness state;

[0050] (2) Impact detection: Determine whether the user's angular velocity data peak value is greater than the preset angular velocity threshold, and whether the user's combined acceleration data peak value is greater than the modified impact acceleration threshold. , if both conditions are met, it means the user is currently in a collision state;

[0051] (3) Posture stability detection: Determine the user's height change Is it greater than the minimum height difference threshold? , and the user's pitch angle change Is it greater than the corrected pitch angle threshold? , and the user's roll angle change Is it greater than the preset roll angle threshold, and is the user's angular velocity variance continuously less than the preset variance threshold for a period of time after the impact? If all four conditions are met, it means that the user is currently in a falling state; wherein, the user's height change It is obtained by calculating the ambient air pressure data.

[0052] Furthermore, the multimodal early warning according to the pre-processed multimodal data in the multi-stage posture detection process further includes:

[0053] First, during the multi-stage posture detection process, a three-dimensional velocity vector is calculated based on the acceleration data in the three-dimensional world coordinate system, and a kinetic energy gradient representing the intensity of the sudden change in motion energy per unit time is calculated based on the three-dimensional velocity vector;

[0054] (1) Three-dimensional velocity vector solution:

[0055] ;

[0056] in, For users at all times The three-dimensional velocity vector of is the acceleration vector in the three-dimensional world coordinate system; For the moment arrive The speed change between For the user at the initial moment The three-dimensional velocity vector of

[0057] (2) Kinetic energy gradient calculation:

[0058] Kinetic energy gradient The basic form:

[0059] ;

[0060] Kinetic energy gradient The expanded form of:

[0061] ;

[0062] in, is the time interval; Quality for users; For users at all times The three-dimensional velocity vector of For users at all times The three-dimensional velocity vector of is the square of the modulus of the velocity change vector; For users at all times The square of the modulus of the three-dimensional velocity vector; For users at all times The square of the modulus of the three-dimensional velocity vector; is the dot product of the two velocity vectors;

[0063] Second, in the multi-stage posture detection process, the user's gait abnormality flag and posture imbalance flag are calculated;

[0064] (1) Calculation of abnormal gait flag: Z-axis acceleration in the three-dimensional world coordinate system Perform FFT analysis to extract its main frequency , when the main frequency And duration When the user's gait abnormality flag is set Set to 1 and trigger gait abnormality warning;

[0065] ;

[0066] (2) Calculation of posture imbalance flag: Monitor the user's posture angle change rate vector , and calculate ,when When the user's posture is unbalanced, the user's posture is unbalanced. Set to 1 and trigger the posture imbalance warning; , To correct the posture imbalance threshold,

[0067] ,

[0068] ,

[0069] in, is the user's pitch angle change rate, is the user's roll angle change rate, is the user's yaw angle change rate, 25° / s is the baseline attitude imbalance threshold, H is the user's height data, in cm; 170 is the base height, in cm; 0.1 is the adjustment coefficient;

[0070] Third, during the multi-stage posture detection process, the user's fall risk assessment value is calculated :

[0071] ,

[0072] in, is the benchmark risk value; Abnormal gait flag The coefficient of Kinetic energy gradient The coefficient of is the pitch angle change The coefficient of It is the user's abnormal gait flag. is the kinetic energy gradient; is the user's pitch angle change; is the error term;

[0073] Fourth, according to the user's fall risk assessment value Conduct graded warnings.

[0074] Furthermore, the step of providing a graded warning based on the user's fall risk assessment value further includes:

[0075] When the user's fall risk assessment value When it is greater than 0.8, it means that the user has gait abnormality and posture imbalance, but no collision has occurred, and the second-level warning is triggered;

[0076] When the user's fall risk assessment value When it is greater than 1.0, it means that the user has abnormal gait and unbalanced posture and is in a falling state, which triggers an emergency warning.

[0077] Furthermore, the step S4 further includes:

[0078] When the second-level warning is triggered, a warning message "Your gait is unstable, please pay attention to safety" is pushed to the user, and a warning text message "The user is at risk of falling" is sent to the emergency contact;

[0079] When an emergency alert is triggered, a call is made to the emergency contact and a text message containing the user's location information is sent for help.

[0080] As a second aspect of the present invention, a system for implementing posture detection and early warning based on multimodal data is provided, which is used to implement the method for implementing posture detection and early warning based on multimodal data as described above. The system for implementing posture detection and early warning based on multimodal data includes:

[0081] A data acquisition layer, configured to acquire multimodal data, wherein the multimodal data includes ambient air pressure data, user acceleration data, user angular velocity data, user magnetic field data, user height data, and user weight data;

[0082] A data processing layer, configured to preprocess the multimodal data to obtain preprocessed multimodal data;

[0083] a core decision layer, configured to perform multi-stage posture detection based on the pre-processed multi-mode data, wherein the multi-stage posture detection includes weightlessness detection, impact detection, and posture stability detection performed sequentially;

[0084] An early warning layer, configured to perform multimodal early warning according to the pre-processed multimodal data during the multi-stage posture detection process;

[0085] The user interaction layer is used to output multi-stage posture detection results and multi-modal warning information.

[0086] The method for realizing posture detection and early warning based on multimodal data provided by the present invention has the following advantages:

[0087] (1) Combining multi-source data from inertial sensors (accelerometers, gyroscopes, magnetometers), barometers, and other sensors, through an innovative multi-stage dynamic threshold judgment algorithm and the introduction of a dynamic compensation mechanism for physiological parameters (height and weight), the system achieves accurate identification and early warning of fall events.

[0088] (2) A dynamic coordinate system is constructed by fusion of spatiotemporal features, solving the problem of random orientation of the device when worn;

[0089] (3) A motion energy analysis model based on the velocity vector change rate and a multimodal logistic regression warning algorithm were proposed to achieve graded prediction of fall risk;

[0090] (4) Compared with traditional fall detection with fixed thresholds or single criteria, the present invention can significantly reduce the false alarm rate and missed alarm rate, provide more personalized, timely and accurate fall risk management, and thus greatly improve the user's safety level and quality of life. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present invention, but do not constitute a limitation of the present invention.

[0092] Figure 1 This is a flow chart of the method for realizing posture detection and early warning based on multimodal data provided by the present invention.

[0093] Figure 2 This is a flow chart of the multi-stage posture detection provided by the present invention.

[0094] Figure 3 This is a flowchart for constructing the three-dimensional world coordinate system provided by the present invention.

[0095] Figure 4 This is a flow chart of the multimodal early warning provided by the present invention.

[0096] Figure 5 This is an architectural diagram of the system for implementing posture detection and early warning based on multimodal data provided by the present invention. DETAILED DESCRIPTION

[0097] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0098] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0099] It should be noted that the terms "first," "second," and the like in the specification and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate for the embodiments of the present invention described herein. In addition, the terms "including," "having," and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatuses.

[0100] In this embodiment, a method for realizing posture detection and early warning based on multi-modal data is provided. Figure 1 The flowchart of the method for realizing posture detection and early warning based on multi-mode data provided by the present invention is as follows: Figure 1 As shown, the method for realizing posture detection and early warning based on multimodal data includes:

[0101] Step S1: Acquire multimodal data, wherein the multimodal data includes ambient air pressure data, user acceleration data, user angular velocity data, user magnetic field data, user height data, and user weight data;

[0102] Preferably, the step S1 further includes:

[0103] Acquire the user's acceleration data in real time through a three-axis accelerometer;

[0104] The user's angular velocity data is obtained in real time through the three-axis gyroscope;

[0105] Acquire the user's magnetic field data in real time through a three-axis magnetometer;

[0106] Obtain ambient air pressure data in real time through the barometer.

[0107] It should be noted that the three-axis accelerometer and three-axis gyroscope are both inertial measurement units (IMUs). The linear acceleration data obtained in real time by the three-axis accelerometer is used for weightlessness detection, impact detection, and motion energy analysis. The angular velocity data obtained in real time by the three-axis gyroscope is used for posture change calculation and impact detection. The magnetic field data obtained in real time by the three-axis magnetometer is used to construct a three-dimensional world coordinate system. The ambient air pressure data obtained by the barometer is used to assist in judging the height change during a fall. The user's height data and the user's weight data are manually entered through the user interface or synchronized from an associated health application.

[0108] Step S2: preprocessing the multimodal data to obtain preprocessed multimodal data;

[0109] Preferably, the step S2 further includes:

[0110] Step S21: Clean, normalize, and perform Kalman filtering on the acceleration data and angular velocity data of the user; wherein the state equation of the Kalman filter is expressed as:

[0111] ;

[0112] ;

[0113] in, For the moment t The system state vector may contain position, velocity, attitude, etc. For the moment The system state vector of F is the state transition matrix, which describes the system state from time At the time How it evolves; B is the control input matrix, which maps the external control input to the state space; For the moment t The control vector, for example, here may be acceleration as input; is the process noise, which represents the unpredictable random disturbance in the model and is usually assumed to be white noise that obeys the normal distribution; For the moment t The measurement vector, namely the user's acceleration data and the user's angular velocity data; is the measurement matrix, which describes how to predict the measurement vector from the system state vector; Measurement noise represents the random error in sensor measurements and is usually assumed to be white noise that obeys a normal distribution. Through a cycle of prediction and correction, the Kalman filter can obtain a more accurate estimate of the true state of the system from noisy sensor data.

[0114] It should be noted that cleaning refers to removing sensor noise and outliers; normalization refers to unifying data of different dimensions into a comparable range; and Kalman filtering refers to real-time filtering of acceleration and angular velocity signals to eliminate noise and ensure the accuracy of subsequent velocity and displacement solutions.

[0115] Step S22: Figure 3 As shown, in order to solve the problem of randomness in the orientation of the device, a three-dimensional world coordinate system is constructed that is independent of the orientation of the device, combining the gravity vector and the magnetic vector. The filtered acceleration data and the filtered angular velocity data are converted to the three-dimensional world coordinate system. The construction process of the three-dimensional world coordinate system is as follows:

[0116] (1) Gravity vector axis determination: normalize the average value of the filtered acceleration data and define it as the axis in the three-dimensional world coordinate system, which serves as the benchmark for subsequent attitude solution and motion analysis;

[0117] ;

[0118] in, is the Z-axis unit vector in the three-dimensional world coordinate system; The average value of the filtered gravity acceleration vector over a period of time. Since the accelerometer mainly measures gravity acceleration when the device is stationary or moving at a constant speed, its average value can be approximated as the direction of gravity. is the modulus of the average value of the gravitational acceleration vector, which is used to normalize the average value of the gravitational acceleration vector to a unit vector;

[0119] (2) Magnetic vector axis determination: Utilize the user's magnetic field data and the Z axis in the three-dimensional world coordinate system to determine the X axis in the three-dimensional world coordinate system, ensuring that the X axis in the three-dimensional world coordinate system is orthogonal to the Z axis;

[0120] ;

[0121] ;

[0122] in, is the three-dimensional world coordinate system axis unit vector; is the magnetic field vector measured by the three-axis magnetometer; is the magnetic field vector The Z-axis unit vector in the three-dimensional world coordinate system projection on; For vector and vector The dot product of is a vector square of the module length; is a vector subtraction operation, the result of which is the same as A vertical vector containing the component of the original magnetic field vector in the horizontal plane, which is normalized to obtain the X-axis.

[0123] Step S3: performing multi-stage posture detection based on the pre-processed multi-mode data, and performing multi-modal early warning based on the pre-processed multi-mode data during the multi-stage posture detection process; wherein the multi-stage posture detection includes weightlessness detection, impact detection, and posture stability detection performed in sequence;

[0124] Preferably, Figure 2As shown, the multi-stage posture detection based on the pre-processed multi-modal data further includes:

[0125] First, the baseline impact acceleration threshold is modified according to the user's weight data to obtain a modified impact acceleration threshold; and the baseline pitch angle threshold is modified according to the user's height data to obtain a modified pitch angle threshold; and the minimum height difference threshold is calculated according to the user's height data; wherein,

[0126] (1) The modified impact acceleration threshold The calculation formula is as follows:

[0127] ;

[0128] Where, is the baseline impact acceleration threshold (usually set to 6g), W is the user's weight in kg; 70 is the baseline weight in kg; 0.3 is the adjustment factor used to control the sensitivity of weight to the threshold;

[0129] It should be noted that the peak impact acceleration generated by a lighter user during an impact may be relatively low, while that generated by a heavier user may be higher. Perform dynamic compensation to avoid missed or false alarms and obtain the corrected impact acceleration threshold For example, for users weighing less than 70kg, the modified impact acceleration threshold is Will be below the baseline impact acceleration threshold Otherwise, it will be higher than the baseline impact acceleration threshold .

[0130] (2) The corrected pitch angle threshold The calculation formula is as follows:

[0131] ;

[0132] Where 75° is the reference pitch angle threshold, H is the user's height data, in cm; 170 is the base height, in cm; -0.2 is the adjustment coefficient;

[0133] It should be noted that when users of different heights fall, the changes in their body posture (such as pitch angle) may be different. H The pitch angle threshold is dynamically adjusted to better suit individual fall characteristics. For example, for taller users, the pitch angle threshold is slightly lowered to more sensitively capture their fall posture.

[0134] (3) The minimum height difference threshold The calculation formula is as follows:

[0135] ;

[0136] Where, H is the user's height data, in cm; 0.55 is the adjustment factor, which indicates the proportion of height loss usually caused by a fall;

[0137] It should be noted that a fall is usually accompanied by a significant drop in the height of the body's center of gravity. , to determine whether ≥ This step is used to further confirm a significant drop in the body's center of gravity during a fall. It is used to assist in determining whether a fall has occurred, as people of different heights fall from different heights.

[0138] Second, firstly, weightlessness detection is performed based on the user's combined acceleration data; after the weightlessness detection is completed, impact detection is performed based on the user's combined acceleration data peak value and angular velocity data peak value; after the impact detection is completed, posture stability detection is performed based on the user's height change, pitch angle change, and roll angle change; wherein,

[0139] (1) Weightlessness detection: Determine whether the user's combined acceleration data SVM is less than a preset weightlessness threshold (e.g., 0.7g) for a period of time (e.g., 80ms). If so, it means that the user is currently in a weightless state;

[0140] It should be noted that the weightlessness false triggering of the sitting behavior of users weighing more than 100 kg is eliminated because when a user weighing more than 100 kg sits down, the combined acceleration data SVM may be less than 0.5g.

[0141] (2) Impact detection: Determine whether the user's angular velocity data peak is greater than a preset angular velocity threshold (e.g., 300° / s), and whether the user's combined acceleration data peak is greater than the corrected impact acceleration threshold. , if both conditions are met, it means the user is currently in a collision state;

[0142] It should be noted that the combined determination of angular velocity and acceleration effectively eliminates interference caused by the wearable device falling. When a wearable device falls, its angular velocity is usually less than 100° / s.

[0143] (3) Posture stability detection: Determine the user's height change Is it greater than the minimum height difference threshold? , and the user's pitch angle change Is it greater than the corrected pitch angle threshold? , and the user's roll angle change Is it greater than a preset roll angle threshold (e.g., 45°), and is the user's angular velocity variance continuously less than a preset variance threshold for a period of time after the impact? If all four conditions are met, it means that the user is currently in a falling state; wherein, the user's height change It is obtained by calculating the ambient air pressure data.

[0144] It should be noted that the pitch angle change is obtained based on the user's angular velocity data and roll angle change .

[0145] Preferably, Figure 4 As shown, the multimodal early warning according to the pre-processed multimodal data in the multi-stage posture detection process also includes:

[0146] First, during the multi-stage posture detection process, a three-dimensional velocity vector is calculated based on the acceleration data in the three-dimensional world coordinate system, and a kinetic energy gradient representing the intensity of the sudden change in motion energy per unit time is calculated based on the three-dimensional velocity vector;

[0147] (1) Three-dimensional velocity vector solution: integration based on acceleration data after Kalman filtering;

[0148] ;

[0149] in, For users at all times The three-dimensional velocity vector of is the acceleration vector in the three-dimensional world coordinate system; For the moment arrive The speed change between For the user at the initial moment The three-dimensional velocity vector of

[0150] (2) Kinetic energy gradient calculation: According to the kinetic energy gradient The size of the pulse can determine the severity of the fall.

[0151] Kinetic energy gradient The basic form:

[0152] ;

[0153] Kinetic energy gradient The expanded form of:

[0154] ;

[0155] in, is the time interval; Quality for users; For users at all times The three-dimensional velocity vector of For users at all times The three-dimensional velocity vector of is the square of the modulus of the velocity change vector; For users at all times The square of the modulus of the three-dimensional velocity vector; For users at all times The square of the modulus of the three-dimensional velocity vector; is the dot product of the two velocity vectors;

[0156] It should be noted that the kinetic energy gradient reflects the rapid increase or decrease in motion energy by quantifying the rapid change in velocity in a short period of time. A violent fall can cause significant velocity changes and energy release, resulting in a high kinetic energy gradient. value, while the kinetic energy gradient of normal sitting or slow falling The value is lower, which effectively distinguishes them.

[0157] Second, in the multi-stage posture detection process, the user's gait abnormality flag and posture imbalance flag are calculated;

[0158] (1) Calculation of abnormal gait flag: Z-axis acceleration in the three-dimensional world coordinate system Perform FFT analysis to extract its main frequency , when the main frequency And duration When the user's gait abnormality flag is set Set to 1 and trigger gait abnormality warning;

[0159] ;

[0160] Among them, in gait analysis, the main frequency It usually corresponds to the user's walking cadence or the most significant periodic frequency in his or her movement. Is a mathematical operator that means "the value that makes the following expression reach the maximum value". After the Fourier transform of az(t), the frequency with the largest amplitude (or energy) is |FFT(az(t))| is the amplitude or modulus obtained by performing fast Fourier transform (FFT) on az(t).

[0161] In a three-dimensional world coordinate system, the Z axis typically represents the vertical direction (aligned with the direction of gravity). Therefore, changes in az(t) can reflect the body's vertical motion, which is closely related to gait (for example, the ups and downs of the body's center of gravity during walking).

[0162] It's important to note that FFT analysis of the Z-axis acceleration az(t) can be used to identify a person's walking cadence. During normal walking, the cadence is typically within a certain range (e.g., around 1.5 Hz). If the cadence drops to a low value (e.g., less than 0.5 Hz) and persists for a period of time, this may indicate an unstable gait, such as slow movement, dragging steps, or unstable footing—all potential signs of a fall.

[0163] (2) Calculation of posture imbalance flag: Monitor the user's posture angle change rate vector , among which, the pitch angle change rate and roll angle change rate are monitored first, and the ,when When the user's posture is unbalanced, the user's posture is unbalanced. Set to 1 and trigger the posture imbalance warning; , To correct the posture imbalance threshold,

[0164] ,

[0165] ,

[0166] in, is the user's pitch angle change rate, is the user's roll angle change rate, is the user's yaw angle change rate, 25° / s is the baseline attitude imbalance threshold, H is the user's height data, in cm; 170 is the base height, in cm; 0.1 is the adjustment coefficient;

[0167] It should be noted that the attitude angle change rate vector Indicates the rate of change of the user's body posture angle relative to time; among them, the pitch angle change rate Indicates the rate of change of the user's body pitch angle over time. The pitch angle usually describes the rotation around the body's lateral axis (X axis), indicating the degree to which the body tilts forward or backward; the roll angle change rate Indicates the rate of change of the user's body roll angle over time. The roll angle usually describes the rotation around the front-back axis (Y axis) of the body, indicating the degree to which the body tilts to the left or right; the yaw angle change rate Indicates the rate of change of the user's body yaw angle over time. The yaw angle usually describes the rotation around the vertical axis of the body (Z axis), indicating the degree to which the body turns left or right.

[0168] It should be noted that taller users may have different posture adjustment strategies or critical points when maintaining balance. By correcting the posture imbalance threshold The formula is used to dynamically compensate the baseline posture imbalance threshold to more accurately capture the individual posture imbalance risk.

[0169] It should be noted that the system predicts fall risk by monitoring the rate of change of the user's posture angle, especially drastic changes in pitch and roll. When the body suddenly loses balance and attempts to recover, the rate of change of the posture angle increases significantly. Combined with the dynamic adjustment of the threshold based on physiological parameters (height), the system can more personalized and accurately identify impending falls.

[0170] Third, during the multi-stage posture detection process, the user's fall risk assessment value is calculated :

[0171] ,

[0172] in, is the benchmark risk value; Abnormal gait flag The coefficient of Kinetic energy gradient The coefficient of is the pitch angle change The coefficient of It is the user's abnormal gait flag. is the kinetic energy gradient; is the user's pitch angle change; is the error term;

[0173] It should be noted that These coefficients are usually learned from a large amount of training data using machine learning algorithms (such as logistic regression). It is usually a binary variable, 1 indicates abnormal gait and 0 indicates normal gait.

[0174] It should be noted that the fall risk assessment value is calculated by linearly combining multiple key features (gait, energy, posture) and learning the weights. This fall risk assessment value It can be used to set different thresholds to achieve graded warning of fall risks, from minor risks to emergency fall events.

[0175] Fourth, according to the user's fall risk assessment value Conduct graded warnings.

[0176] Furthermore, the step of providing a graded warning based on the user's fall risk assessment value further includes:

[0177] When the user's fall risk assessment value When it is greater than 0.8, it means that the user has gait abnormality and posture imbalance, but no collision has occurred, and the second-level warning is triggered;

[0178] When the user's fall risk assessment value When it is greater than 1.0, it means that the user has abnormal gait and unbalanced posture and is in a falling state, which triggers an emergency warning.

[0179] Step S4: Output multi-stage posture detection results and multi-modal warning information.

[0180] Preferably, the step S4 further includes:

[0181] When the second-level warning is triggered, a warning message "Your gait is unstable, please pay attention to safety" is pushed to the user's APP, and a warning text message "The user is at risk of falling" is sent to the emergency contact;

[0182] When an emergency alert is triggered, a call is made to the emergency contact and a text message containing the user's location information is sent for help.

[0183] The present invention is suitable for scenarios requiring high-precision, low-false-alarm fall detection and warning. Its application areas include but are not limited to the following categories: (1) Smart wearable devices: It can be integrated into various portable devices such as smart watches, bracelets, smart phones, smart shoes, etc., to provide individual users with real-time fall detection and emergency assistance functions. (2) Elderly care and home safety: As a core component of home-based elderly care and smart community elderly care solutions, it provides fall risk monitoring, fall event alarms and linkage for the elderly, reducing accidental risks. (3) Medical care and rehabilitation: It can be used to monitor the gait stability of high-risk groups (such as Parkinson's patients and stroke recovery patients), assess fall risks, and assist in evaluating the effectiveness of rehabilitation training. (4) Sports and occupational safety: In high-risk sports (such as mountaineering and skiing) or special occupations (such as high-altitude operations and miners), it provides real-time warnings of falls or accidental falls to ensure personal safety. (5) Smart cities and public safety: Combined with smart street lights, public place monitoring systems, etc., a regional fall monitoring network is constructed to enhance the city's emergency response capabilities.

[0184] This invention can not only effectively solve the pain points of existing technologies and improve user experience, but also give rise to value-added services based on data analysis and risk management, and has broad market prospects and social benefits.

[0185] The method for posture detection and early warning based on multimodal data provided by this invention uses multiple sensors built into a user's smart wearable device (such as a mobile phone or watch) to collect real-time environmental information and human motion data. After data cleaning, normalization, feature extraction, and fusion processing, this data is fed into an algorithm processing module. Based on this data, the algorithm processing module employs a physiological parameter compensation mechanism, a multi-stage dynamic threshold decision algorithm, a spatiotemporal feature fusion strategy, a motion energy analysis model based on the velocity vector change rate, and a multimodal early warning algorithm to accurately identify fall events and provide graded early warnings for fall risks. Ultimately, this method enables high-precision, low-false-alarm fall detection and early risk management.

[0186] The following describes in detail the method for realizing posture detection and early warning based on multimodal data provided by the present invention using two specific cases.

[0187] Case 1: Smart Bracelet Fall Detection and Graded Warning

[0188] (1) Data collection: The user wears a smart bracelet with a built-in accelerometer, gyroscope, magnetometer, and barometer. The smart bracelet collects three-axis acceleration, three-axis angular velocity, and ambient air pressure data in real time. At the same time, the user enters a height of 160 cm and a weight of 60 kg in the APP when using it for the first time.

[0189] (2) Data processing and coordinate system construction: The microcontroller built into the wristband performs Kalman filtering to denoise the raw sensor data. Subsequently, the world coordinate system is dynamically constructed based on the real-time average of the accelerometer data and the magnetometer data. For example, when the wristband is rotated from the front to the back of the wrist, its own coordinate system will change, but after the world coordinate system conversion, the calculated human posture angle remains stable.

[0190] (3) Core judgment - physiological parameter compensation and multi-stage judgment:

[0191] Physiological parameter compensation: Dynamically adjust the impact acceleration threshold based on the user's weight of 60kg :

[0192] ;

[0193] Adjust the pitch angle threshold based on the user's height of 160cm :

[0194] ;

[0195] Height difference threshold :

[0196] ;

[0197] Posture imbalance threshold :

[0198] ;

[0199] Weightlessness detection: combined acceleration Support Vector Machine The value is less than 0.7g for 80ms. At the same time, the false triggering of the sitting behavior of users weighing more than 100kg is eliminated because when sitting down, Support Vector Machine Probably less than 0.5g.

[0200] Impact detection: Then, the bracelet detects the angular velocity (greater than 300° / s) and combined acceleration Support Vector Machine =7.0g (greater than =6.4g). Since both high angular velocity and high acceleration are met, the system determines it as a collision. Support Vector Machine Up to 7.0g, and If the speed is only 80° / s, the system will determine that the device has been dropped rather than fallen.

[0201] Attitude stability detection: After the collision, the system detects the change in pitch angle =78° (greater than 72°) and roll angle changes =50° (greater than 45°), and at the same time, the altitude change calculated by the barometer data is 100cm (greater than =88cm). At the same time, the angular velocity variance was detected to be less than 5° / s within 5s after the impact. 2 And the user did not cancel manually, and the fall event was finally confirmed.

[0202] (4) Multimodal fusion and graded warning:

[0203] Gait instability prediction: 3 seconds before impact, the user experiences shaking. The system detects that the FFT main frequency of the world coordinate system Z-axis acceleration signal drops suddenly from 1.5Hz during normal walking to 0.Hz and persists for 2 seconds, triggering the gait abnormality flag Step1=1.

[0204] Posture imbalance prediction: 1s before the impact, the user's body shakes violently in an attempt to restore balance, and the posture angle change rate vector The absolute value of the projection in the pitch / roll direction reaches 30° / s, which is greater than the corrected threshold of 25.5° / s, triggering an attitude imbalance signal.

[0205] Kinetic energy gradient discrimination: At the moment of impact, the system calculates the kinetic energy gradient =200J / s, much greater than 150J / s, confirming a violent fall.

[0206] Logistic regression and graded warning: Assume that the coefficients of the logistic regression model after training are .

[0207] When gait instability and postural imbalance occur, but no impact occurs (for example, the user is about to fall and is caught), calculate: ,

[0208] ,

[0209] at this time >0.8, triggering a Level 2 warning, pushing a message to the user's APP saying "Your gait is unstable, please pay attention to safety" and sending a text message to emergency contacts saying "The user is at risk of falling."

[0210] When a fall event is confirmed, calculate: ,

[0211] ,

[0212] at this time >1.0, triggers an emergency warning, immediately calls the emergency contact and sends a text message with location information for help.

[0213] User interaction: Warning information is notified to users and emergency contacts through various means such as bracelet vibration, APP pop-up window, phone call, SMS, etc.

[0214] Case 2: Fall Detection for Smart Devices Without Barometers

[0215] This case is similar to Case 1, but on devices without a barometer, the system will perform FFT analysis on the Z-axis acceleration signal of the world coordinate system to detect a sudden drop or disappearance of the cadence to assist in determining the height change during a fall, rather than relying on the barometer data. For example, when the main frequency is detected f <0.5Hz and continuous When it is greater than 1.5s, it is used as one of the auxiliary criteria for falls.

[0216] As another embodiment of the present invention, Figure 5 As shown, a system for realizing posture detection and early warning based on multi-mode data is provided, wherein the system for realizing posture detection and early warning based on multi-mode data includes:

[0217] A data acquisition layer, configured to acquire multimodal data, wherein the multimodal data includes ambient air pressure data, user acceleration data, user angular velocity data, user magnetic field data, user height data, and user weight data;

[0218] A data processing layer, configured to preprocess the multimodal data to obtain preprocessed multimodal data;

[0219] a core decision layer, configured to perform multi-stage posture detection based on the pre-processed multi-mode data, wherein the multi-stage posture detection includes weightlessness detection, impact detection, and posture stability detection performed sequentially;

[0220] An early warning layer, configured to perform multimodal early warning according to the pre-processed multimodal data during the multi-stage posture detection process;

[0221] The user interaction layer is used to output multi-stage posture detection results and multi-modal warning information.

[0222] Specifically, the user interaction layer primarily provides users with a visual interface and warning information. This layer includes a visual interface and warning notifications. The visual interface displays real-time motion data, risk levels, and historical fall records. Warning notifications send graded warning information to users or emergency contacts via mobile app push, text messages, and phone calls.

[0223] The main functions of the present invention are as follows: (1) Personalized threshold adaptation: The system dynamically adjusts the key threshold of fall detection according to the user's height and weight to ensure the personalization and accuracy of detection. (2) Multi-sensor fusion and coordinate system independence: The dynamic world coordinate system is constructed by gravity vector and magnetic vector to eliminate the influence of the device wearing orientation on the interpretation of sensor data and improve the robustness of the algorithm. (3) Multi-stage logical judgment to reduce false alarms: The characteristics of the three stages of weightlessness, impact, and posture stability are combined for layer-by-layer verification, especially the "angular velocity + acceleration joint judgment" to effectively distinguish between real falls and daily activities or device drops. (4) Energy dimension judgment to improve accuracy: The physical quantity "kinetic energy gradient" is introduced to capture the severity of the fall more fundamentally from the perspective of energy change, thereby improving the judgment accuracy. (5) Early risk prediction and graded warning: Through the gait instability and posture imbalance model, potential risks are identified before the fall occurs, and risk classification is achieved by combining the logistic regression model, providing differentiated warning responses, and realizing the transformation from "post-event alarm" to "pre-event prevention". (6) Environmental adaptation and intelligent compensation: The barometer assists in verifying altitude changes, or FFT analysis of cadence disappearance is used as an alternative to enhance the system's applicability under different hardware configurations.

[0224] The innovations of the present invention are as follows: (1) Dynamic compensation mechanism for physiological parameters: According to the physiological differences of users of different heights and weights, key thresholds such as impact acceleration, pitch angle, height difference and attitude angle change rate are dynamically adjusted, which significantly improves the personalized adaptability and accuracy of fall detection and reduces false alarms and missed alarms. (2) Multi-stage joint judgment algorithm: For the first time, the three-stage series judgment logic of weightlessness, impact and attitude stability is proposed and applied, especially the "joint judgment of angular velocity and acceleration", which effectively distinguishes real falls from non-fall events such as device dropping and sitting down, and greatly improves the robustness of the system in complex scenarios. (3) Construction of dynamic world coordinate system: Using accelerometer and magnetometer data, the gravity vector and magnetic vector are used to determine the axis, and a world coordinate system that is independent of the device's own orientation is dynamically constructed, which solves the problem of inaccurate posture calculation caused by the random orientation of handheld or worn devices and enhances the universality of the algorithm. (4) Fall judgment based on kinetic energy gradient: The "kinetic energy gradient" is innovatively introduced. "As a fall criterion, by quantifying the sudden change intensity of motion energy per unit time, it can distinguish between normal sitting and violent falls from the physical essence, providing a novel and more discriminative dimension for fall identification. (5) Multimodal fusion graded warning algorithm: Combining multiple features such as gait instability, posture imbalance, kinetic energy gradient, etc., a comprehensive judgment is made through a multi-conditional logistic regression model to achieve early prediction of fall risks (first / second level warning), and provide differentiated responses according to the risk level, realizing a paradigm shift from "post-event alarm" to "pre-event warning", which has significant social and commercial value.

[0225] It will be understood that the above embodiments are merely exemplary embodiments for illustrating the principles of the present invention, and the present invention is not limited thereto. Those skilled in the art will appreciate that various modifications and improvements can be made without departing from the spirit and substance of the present invention, and such modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for realizing posture detection and early warning based on multimodal data, characterized in that: The method for realizing posture detection and early warning based on multimodal data includes: Step S1: Acquire multimodal data, wherein the multimodal data includes ambient air pressure data, user acceleration data, user angular velocity data, user magnetic field data, user height data, and user weight data; Step S2: preprocessing the multimodal data to obtain preprocessed multimodal data; Step S3: performing multi-stage posture detection based on the pre-processed multi-mode data, and performing multi-modal early warning based on the pre-processed multi-mode data during the multi-stage posture detection process; wherein the multi-stage posture detection includes weightlessness detection, impact detection, and posture stability detection performed in sequence; Step S4: outputting multi-stage posture detection results and multi-modal warning information; The multi-stage posture detection according to the pre-processed multimodal data further includes: The baseline impact acceleration threshold is modified according to the user's weight data to obtain a modified impact acceleration threshold; the baseline pitch angle threshold is modified according to the user's height data to obtain a modified pitch angle threshold; and the minimum height difference threshold is calculated according to the user's height data; wherein, (1) The modified impact acceleration threshold The calculation formula is as follows: ， Where, is the baseline impact acceleration threshold; The weight data of the user, in units of ; is the base weight, in units of ; is the adjustment factor; (2) The corrected pitch angle threshold The calculation formula is as follows: ， Where, is the reference pitch angle threshold; The height data of the user, in units of ; is the base height, in units of ; is the adjustment factor; (3) The minimum height difference threshold The calculation formula is as follows: ， Where, The height data of the user, in units of ; The multi-modal early warning according to the pre-processed multi-modal data during the multi-stage posture detection process further includes: First, during the multi-stage posture detection process, the three-dimensional velocity vector is calculated based on the acceleration data in the three-dimensional world coordinate system. The kinetic energy gradient, which represents the intensity of the sudden change in motion energy per unit time, is calculated based on the three-dimensional velocity vector. (1) Three-dimensional velocity vector solution: ， in, For users at all times The three-dimensional velocity vector of is the acceleration vector in the three-dimensional world coordinate system; For the moment arrive The speed change between For the user at the initial moment The three-dimensional velocity vector of (2) Kinetic energy gradient calculation: Kinetic energy gradient The basic form: ， Kinetic energy gradient The expanded form of: ， in, is the time interval; Quality for users; For users at all times The three-dimensional velocity vector of For users at all times The three-dimensional velocity vector of is the square of the modulus of the velocity change vector; For users at all times The square of the modulus of the three-dimensional velocity vector; For users at all times The square of the modulus of the three-dimensional velocity vector; is the dot product of the two velocity vectors; Second, in the multi-stage posture detection process, the user's gait abnormality flag and posture imbalance flag are calculated; (1) Calculation of abnormal gait flag: Axis acceleration conduct Analyze and extract its main frequency , when the main frequency And duration When the user's gait abnormality flag is set Set to 1 and trigger gait abnormality warning; ， (2) Calculation of posture imbalance flag: Monitor the user's posture angle change rate vector , and calculate ,when When the user's posture is unbalanced, the user's posture is unbalanced. Set to 1 and trigger the posture imbalance warning; , To correct the posture imbalance threshold, ， ， in, is the user's pitch angle change rate, is the user's roll angle change rate, is the user's yaw angle change rate, is the baseline posture imbalance threshold; The user's height data, in units of ; is the base height, in units of ; is the adjustment factor; Third, during the multi-stage posture detection process, the user's fall risk assessment value is calculated : ， in, is the benchmark risk value; Abnormal gait flag The coefficient of Kinetic energy gradient The coefficient of is the pitch angle change The coefficient of It is the user's abnormal gait flag. is the kinetic energy gradient; is the user's pitch angle change; is the error term; Fourth, according to the user's fall risk assessment value Conduct graded warnings.

2. The method for realizing posture detection and early warning based on multimodal data according to claim 1, characterized in that: The step S1 further includes: Acquire the user's acceleration data in real time through a three-axis accelerometer; The user's angular velocity data is obtained in real time through the three-axis gyroscope; Acquire the user's magnetic field data in real time through a three-axis magnetometer; Obtain ambient air pressure data in real time through the barometer.

3. The method for realizing posture detection and early warning based on multimodal data according to claim 1, characterized in that: The step S2 further includes: Step S21: Clean, normalize, and perform Kalman filtering on the acceleration data and angular velocity data of the user; wherein the state equation of the Kalman filter is expressed as: ； ， in, is the system state vector at time t; For the moment The system state vector; F is the state transfer matrix; B is the control input matrix; is the control vector at time t; is the process noise; is the measurement vector at time t, i.e., the user’s acceleration data and the user’s angular velocity data; is the measurement matrix; To measure noise; Step S22: Constructing a three-dimensional world coordinate system by combining the gravity vector and the magnetic vector, and converting the filtered acceleration data and the filtered angular velocity data into the three-dimensional world coordinate system; wherein the construction process of the three-dimensional world coordinate system is as follows: (1) Gravity vector axis: normalize the average value of the filtered acceleration data to define it as axis; ， in, is the three-dimensional world coordinate system axis unit vector; is the average value of the filtered gravity acceleration vector over a period of time; is the modulus of the average value of the gravitational acceleration vector, which is used to normalize the average value of the gravitational acceleration vector to a unit vector; (2) Magnetic vector axis determination: Utilize the user's magnetic field data and combine it with the three-dimensional world coordinate system. Axis, determine the three-dimensional world coordinate system axis, ensuring that the three-dimensional world coordinate system Axis and Axis orthogonal; ， ， in, is the three-dimensional world coordinate system axis unit vector; is the magnetic field vector measured by the three-axis magnetometer; is the magnetic field vector In the three-dimensional world coordinate system Axis unit vector projection on; For vector and vector The dot product of is a vector The square of the module length.

4. The method for realizing posture detection and early warning based on multimodal data according to claim 3, characterized in that: The multi-stage posture detection according to the pre-processed multi-modal data further includes: First, weightlessness detection is performed based on the user's combined acceleration data; after the weightlessness detection is completed, impact detection is performed based on the user's combined acceleration data peak value and angular velocity data peak value; after the impact detection is completed, posture stability detection is performed based on the user's height change, pitch angle change, and roll angle change; wherein, (1) Weightlessness detection: Determine whether the user's combined acceleration data is less than the preset weightlessness threshold for a period of time. If so, it means that the user is currently in a weightlessness state; (2) Impact detection: Determine whether the user's angular velocity data peak is greater than the preset angular velocity threshold, and whether the user's combined acceleration data peak is greater than the modified impact acceleration threshold , if both conditions are met, it means the user is currently in a collision state; (3) Posture stability detection: Determine the user's height change Is H greater than the minimum height difference threshold? , and the user's pitch angle change Is it greater than the corrected pitch angle threshold? , and the user's roll angle change Is it greater than the preset roll angle threshold, and is the user's angular velocity variance continuously less than the preset variance threshold for a period of time after the impact? If all four conditions are met, it means that the user is currently in a falling state; wherein, the user's height change H is obtained by calculating the ambient air pressure data.

5. The method for realizing posture detection and early warning based on multimodal data according to claim 1, characterized in that: The fall risk assessment value of the user The graded warning also includes: when , indicating that the user has gait abnormality and posture imbalance, but no collision has occurred, triggering the second-level warning; when , indicating that the user has abnormal gait and unbalanced posture and is in a falling state, which triggers an emergency warning.

6. The method for realizing posture detection and early warning based on multimodal data according to claim 5, characterized in that: The step S4 further includes: After triggering the second level warning, the user Push a warning message saying "Your gait is unstable, please pay attention to safety" and send a warning SMS message to emergency contacts saying "The user is at risk of falling"; When an emergency alert is triggered, a call is made to the emergency contact and a text message containing the user's location information is sent for help.

7. A system for realizing posture detection and early warning based on multimodal data, for realizing the method for realizing posture detection and early warning based on multimodal data according to any one of claims 1 to 6, characterized in that: The system for realizing posture detection and early warning based on multi-mode data includes: A data acquisition layer, configured to acquire multimodal data, wherein the multimodal data includes ambient air pressure data, user acceleration data, user angular velocity data, user magnetic field data, user height data, and user weight data; A data processing layer, configured to preprocess the multimodal data to obtain preprocessed multimodal data; a core decision layer, configured to perform multi-stage posture detection based on the pre-processed multi-mode data, wherein the multi-stage posture detection includes weightlessness detection, impact detection, and posture stability detection performed sequentially; An early warning layer, configured to perform multimodal early warning according to the pre-processed multimodal data during the multi-stage posture detection process; The user interaction layer is used to output multi-stage posture detection results and multi-modal warning information.

Images