Intelligent wearable pelvis posture real-time correction method and device

Abstract

The application discloses a kind of intelligent wearing type pelvis posture real-time correction method and equipment, it is related to intelligent wearing device technical field, including: in intelligent wearing device integration pelvis posture monitoring sensor, real-time acquisition user's pelvis movement data;Pelvis movement data is handled and analyzed, and the feature parameter set related to pelvis posture is extracted;The feature parameter set extracted is compared and analyzed with the normal pelvis posture stored in advance;Using machine learning algorithm and pattern recognition technique, whether the current pelvis posture is in normal range is judged;If pelvis posture anomaly is detected, then the abnormal type of pelvis posture and the abnormal degree are judged;According to the judgment result, different posture abnormality warning signals are sent to user.The application can capture the subtle movement characteristics of pelvis in real time, accurately identify and grade determine various abnormal types in combination with advanced machine learning algorithm, and realize personalized reminding through dynamic adjustment of feedback parameters.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent wearable devices, and more particularly to an intelligent wearable pelvis posture real-time correction method and device. BACKGROUND

[0002] The pelvis occupies a central position in the human body structure, and its structure and function have evolved continuously in the long evolution of mankind. In modern society, the pelvis is particularly important for maintaining the normal physiological function and posture balance of the body. However, with the great changes in modern lifestyle, people's daily activity patterns have changed significantly, and bad habits such as sitting at a desk for a long time and using electronic devices for a long time are becoming increasingly common. These behaviors have brought unprecedented challenges to the stability of the pelvis, leading to a sharp rise in the incidence of abnormal pelvis posture year by year.

[0003] According to statistics from the World Health Organization, about 42% of adults worldwide have different degrees of forward or backward problems of the pelvis. Abnormal pelvis posture not only affects the appearance of the body, but also causes a series of serious health problems such as back pain and scoliosis. These diseases not only bring physical pain to patients, but also cause great inconvenience to their daily life and work, and have gradually become a global public health problem.

[0004] Traditional pelvis correction methods mainly rely on manual intervention or static correction devices. Manual intervention requires professional medical personnel to guide and operate, which not only has high cost, but also is difficult to achieve real-time, continuous monitoring and correction. Static correction devices can only be used in specific environments and cannot monitor and adjust the dynamic pelvis posture of users in daily life, making it difficult to meet the needs of modern people for convenient and efficient health management.

[0005] In the prior art, some wearable devices attempt to monitor the pelvis posture through a single sensor, but this approach has obvious limitations. The single sensor can only collect limited data dimensions, and cannot fully and accurately capture the various subtle changes in the pelvis. At the same time, its ability to identify abnormal types of pelvis posture is weak, and it is difficult to accurately distinguish between forward, backward, lateral tilt and other types of abnormalities, and to accurately judge the degree of abnormality.

[0006] Therefore, how to provide an intelligent wearable pelvis posture real-time correction method and device that can capture subtle pelvis movement characteristics in real time, combine advanced machine learning algorithms to accurately identify and grade various abnormal types, and achieve personalized reminders through dynamic adjustment of feedback parameters is a problem that needs to be solved by those skilled in the art. SUMMARY

[0007] Therefore, the application provides a smart wearable pelvis posture real-time correction method and device, which can capture the subtle motion characteristics of the pelvis in real time and comprehensively, accurately identify and grade determine abnormal types such as pelvis forward inclination, backward inclination and lateral inclination by using a machine learning algorithm and a pattern recognition technology, and send personalized posture abnormality reminding signals to the user by dynamically adjusting vibration feedback parameters, so as to guide the user to timely and independently adjust the posture, thereby effectively preventing and correcting the pelvis posture abnormality and improving the physical health of the user.

[0008] In order to achieve the above-mentioned purpose, the application adopts the following technical scheme: a smart wearable pelvis posture real-time correction method, comprising:

[0009] A pelvis posture monitoring sensor is integrated in the smart wearable device to collect pelvis motion data of the user in real time;

[0010] The pelvis motion data is processed and analyzed to extract a feature parameter set related to the pelvis posture;

[0011] The extracted feature parameter set is compared and analyzed with a pre-stored normal pelvis posture;

[0012] A machine learning algorithm and a pattern recognition technology are used to determine whether the current pelvis posture is within a normal range;

[0013] If the pelvis posture abnormality is detected, the abnormal type and the abnormal degree of the pelvis posture are determined; and different posture abnormality reminding signals are sent to the user according to the determination result.

[0014] Preferably, the real-time pelvis motion data and the correction execution situation are transmitted and stored to a user terminal device.

[0015] Preferably, the extracted feature parameter set related to the pelvis posture comprises:

[0016] Based on a feature pyramid network, an improved side fusion network is introduced to establish a pelvis posture feature extraction network;

[0017] The feature parameter set related to the pelvis posture is extracted by the pelvis posture feature extraction network.

[0018] Preferably, based on the feature pyramid network, the improved side fusion network is introduced to establish the pelvis posture feature extraction network, which comprises:

[0019] The pelvis motion data is processed in stages by using the feature pyramid network; in each stage, a convolution operation is performed, followed by a pooling operation, to obtain feature parameters {C1, C2, C3, C4, C5} of different stages;

[0020] The up-sampling method is used to amplify the top layer feature parameter {C5} to make its dimension size consistent with the feature parameter of the previous stage in a top-down fusion path;

[0021] The amplified feature parameter is added to the corresponding element of the feature parameter of the previous stage to obtain the fusion feature parameter {P2, P3, P4, P5};

[0022] The side fusion network structure is introduced, and {N1} is directly equal to {C1} and {N5} is directly equal to {C5} between the feature layers of the same stage and the same scale, so that they remain unchanged;

[0023] On this basis, the skip connection path from {C2, C3, C4} to {N2, N3, N4} is added; the feature parameters of {C2, C3, C4} are fused with the feature parameters of {P2, P3, P4} fused from the top through the skip connection path to generate the feature parameter set {N1, N2, N3, N4, N5} containing the feature information of the current layer and the fused feature information of the top layer.

[0024] Preferably, the machine learning algorithm and pattern recognition technology are used to determine whether the current pelvic posture is within the normal range, including:

[0025] A feature parameter in the feature parameter set is selected to construct a new optimization problem, and the new optimization problem is used to replace the basic optimization problem in the traditional SVM model;

[0026] The sigmoid kernel function is selected as the kernel function of the traditional SVM model to obtain a preliminary improved SVM model;

[0027] The hyperparameters of the preliminary improved SVM model are optimized, and the optimal weight vector ω and bias term b are updated to obtain the trained improved SVM model;

[0028] Based on the mean absolute error and the root mean square error, the values of the correlation coefficient R and the relative error σ are obtained, and R and σ are used to evaluate the performance of the trained improved SVM model. At the same time, the regularization parameter C is adjusted to control the complexity of the trained improved SVM model;

[0029] The evaluated improved SVM model is used for new data in the feature parameter set, and the value of ω·x i +b in the evaluated improved SVM model is calculated, and the data is assigned to the category with a calculated value greater than 0.05 to obtain the classified improved SVM model;

[0030] The other feature parameters in the feature parameter set are repeated to output and update the weight and bias of the preliminary improved SVM model, complete the learning of the preliminary improved SVM model, and obtain the final improved SVM prediction model.

[0031] Preferably, the preliminary improved SVM model is super parameter optimized to update its optimal weight vector ω and bias term b to obtain a trained improved SVM model, including:

[0032] A plurality of super parameters are randomly selected in the preliminary improved SVM model to form a first generation data set;

[0033] The fitness function is used to calculate the fitness of each individual in the first generation data set, and it is judged whether the individual meets the termination condition at this time. If it meets, the optimal weight vector ω and bias term b of the preliminary improved SVM model are output and updated;

[0034] If not, the probability of each individual being selected in the next selection is calculated according to the obtained fitness and the probability formula of individual i being selected;

[0035] According to the calculated probability, M individuals with large probability are selected to form a second generation data set; wherein, a certain data can be repeatedly selected, and M < N;

[0036] The data in the second generation data set is operated based on the mating probability pc to generate N-M data, and the N-M data is subjected to mutation operation to obtain a mutated N-M data set. L data are generated in the second generation data set in the form of copying, and the L data are merged with the mutated N-M data set to generate a third generation data set;

[0037] The third generation data set is used to judge whether the individual meets the termination condition at this time. If it meets, the optimal weight vector ω and bias term b of the preliminary improved SVM model are output and updated to realize super parameter optimization of the preliminary improved SVM model and obtain a trained improved SVM model.

[0038] Preferably, the performance of the final improved SVM prediction model is measured by calculating the values of correlation coefficient R and relative error σ based on the mean absolute error and root mean square error, and the performance of the final improved SVM prediction model is measured by the values of correlation coefficient R and relative error σ.

[0039] Preferably, an intelligent wearable pelvis posture real-time correction device, comprising:

[0040] The intelligent waistband is made of elastic material, and the intelligent waistband is internally provided with a pelvis posture monitoring sensor, a microprocessor, a Bluetooth module and a battery;

[0041] The mobile terminal is connected with the intelligent waistband through Bluetooth, and is used for receiving and storing data transmitted by the intelligent waistband.

[0042] The vibration motor is installed in the inner ring of the intelligent waistband, when the abnormal posture of the pelvis is detected, the vibration motor is controlled to generate vibration with different frequency and intensity according to different types of abnormality, so as to remind the user to adjust the posture in time.

[0043] Preferably, the pelvis posture monitoring sensor is used to collect the motion, angle and pressure data of the pelvis in real time and transmit them to the microprocessor for analysis and processing.

[0044] Compared with the prior art, the technical scheme can provide a kind of intelligent wearing type pelvis posture real-time correction method and equipment, related to intelligent wearing equipment technical field, including: in the intelligent wearing equipment, integrate pelvis posture monitoring sensor, real-time acquisition user's pelvis motion data;Pelvis motion data is handled and analyzed, and the feature parameter set related to pelvis posture is extracted;The feature parameter set extracted is compared and analyzed with the normal pelvis posture stored in advance;Machine learning algorithm and pattern recognition technology are used to judge whether the current pelvis posture is within the normal range;If the pelvis posture is abnormal, the abnormal type and abnormal degree of the pelvis posture are judged;According to the judgment result, different posture abnormality warning signals are sent to the user.The present application can capture the fine motion characteristics of the pelvis in real time, accurately identify and grade determine various abnormal types by combining advanced machine learning algorithms, and realize personalized reminding by dynamically adjusting feedback parameters. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.

[0046] Figure 1 A flow chart of a smart wearable pelvis posture real-time correction method provided by the present application.

[0047] Figure 2 A structural schematic diagram of a smart wearable pelvis posture real-time correction device provided by the present application. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0049] The embodiment of the application discloses a kind of intelligent wear type pelvis posture real-time correction methods, as shown in Figure 1 Including:

[0050] Pelvis posture monitoring sensor is integrated in intelligent wearable device, and the pelvis posture monitoring sensor includes acceleration sensor, gyroscope sensor and pressure sensor, and the pelvis movement data of user is collected in real time;Including acceleration, angular velocity and the like information, these sensors can work at higher sampling frequency, to ensure that any subtle action change of pelvis can be captured, to provide rich, accurate data basis for subsequent analysis and judgment;

[0051] Pelvis movement data is processed and analyzed, and the feature parameter set related to pelvis posture is extracted;

[0052] The feature parameter set extracted is compared and analyzed with the normal pelvis posture stored in advance;

[0053] Machine learning algorithm and pattern recognition technology are used to judge whether the current pelvis posture is within normal range;

[0054] If pelvis posture anomaly is detected, the abnormal type and the abnormal degree of pelvis posture are judged;The abnormal type of pelvis posture includes forward inclination, backward inclination, lateral inclination and the like, and the abnormal degree of pelvis posture is judged according to monitored forward inclination angle, backward inclination angle and lateral inclination angle;

[0055] According to the judgment result, different posture abnormality reminding signals are sent to user;

[0056] According to the judgment result, corresponding control instruction is sent, and different posture abnormality reminding signals are sent to user by adjusting vibration intensity and frequency of vibration motor, to guide user to adjust posture independently.

[0057] Specifically, real-time pelvis movement data and correction execution situation are transmitted and stored to user terminal device. User can intuitively view own pelvis posture change history, correction record and current posture state and the like information through APP.

[0058] Specifically, the pelvis movement data is processed, including filtering, noise reduction and coordinate conversion processing, to remove noise and interference in data and improve data quality.

[0059] The feature parameter set related to pelvis posture is extracted, including:

[0060] Based on feature pyramid network, improved side fusion network is introduced, and pelvis posture feature extraction network is established;

[0061] The feature parameter set related to pelvis posture is extracted through the pelvis posture feature extraction network.

[0062] The pelvic posture feature extraction network extracts feature parameters closely related to the pelvic posture from the preprocessed data, such as the inclination angle, rotation angle, movement speed, and acceleration change rate of the pelvis. These feature parameters can accurately reflect the posture information of the pelvis and provide key basis for subsequent abnormality judgment.

[0063] Specifically, based on the feature pyramid network, an improved side fusion network is introduced to establish the pelvic posture feature extraction network, including:

[0064] The feature pyramid network is used to process the pelvic movement data in stages; in each stage, convolution operation is performed in turn to extract local features in the data; then, pooling operation is performed to reduce the data dimension and highlight the main features, thereby obtaining feature parameters {C1, C2, C3, C4, C5} at different stages; the feature parameter {C5} is located at the top layer of the feature pyramid network;

[0065] A top-down fusion path is used to use the up-sampling method to enlarge the feature parameter {C5} at the top layer, so that its dimension size is consistent with that of the feature parameter at the previous stage;

[0066] The enlarged feature parameter and the feature parameter at the previous stage are added to obtain the fusion feature parameter {P2, P3, P4, P5}; this process realizes the fusion of feature parameters at different scales and enriches the feature information;

[0067] The side fusion network structure is introduced, and in the same stage and at the same scale of the feature layer, {N1} is directly equal to {C1}, and {N5} is directly equal to {C5}, which remains unchanged;

[0068] On this basis, a jump connection path from {C2, C3, C4} to {N2, N3, N4} is added; through the jump connection path, the feature parameters of {C2, C3, C4} are side-fused with the feature parameters {P2, P3, P4} after top-down fusion to generate the {N1, N2, N3, N4, N5} feature parameter set containing detailed feature information of the current layer and fused top layer feature information.

[0069] These feature parameters will provide key data basis for subsequent accurate judgment of the pelvic posture.

[0070] In the convolution operation of the feature pyramid network, a convolution kernel with a size of 3x3 and a step size of 1 is used; the pooling operation uses a pooling kernel with a size of 2x2 and a step size of 2; the up-sampling multiple in the top-down fusion path is 2.

[0071] Specifically, machine learning algorithms and pattern recognition techniques are used to determine whether the current pelvic posture is within the normal range, including:

[0072] selecting one of the feature parameters in the feature parameter set to construct a new optimization problem, and replacing the basic optimization problem in the traditional SVM model with the new optimization problem;

[0073] selecting a sigmoid kernel function as the kernel function of the traditional SVM model to obtain a preliminary improved SVM model;

[0074] optimizing the hyperparameters of the preliminary improved SVM model, updating the optimal weight vector ω and the bias term b thereof to obtain a trained improved SVM model;

[0075] based on the mean absolute error and the root mean square error, obtaining the values of the correlation coefficient R and the relative error σ, using R and σ to evaluate the performance of the trained improved SVM model, and adjusting the regularization parameter C to control the complexity of the trained improved SVM model;

[0076] using the evaluated improved SVM model for new data in the feature parameter set, and calculating the value of ω·x i +b in the evaluated improved SVM model, assigning the data to a category with a calculated value greater than 0.05 to obtain a classified improved SVM model;

[0077] repeating the above operations for other feature parameters in the feature parameter set, outputting and updating the weights and biases of the preliminary improved SVM model, completing the learning of the preliminary improved SVM model, and obtaining a final improved SVM prediction model.

[0078] inputting the pelvis posture feature parameter set obtained by real-time acquisition and processing into the final improved SVM prediction model, and the model will output a judgment result of whether the current pelvis posture is normal.

[0079] If the pelvis posture is detected to be abnormal, the type (forward inclination, backward inclination, lateral inclination, etc.) and degree of abnormality are determined according to the monitored forward inclination angle, backward inclination angle and lateral inclination angle.

[0080] According to the judgment result, different posture abnormality reminding signals are sent to the user, the user is guided to adjust the posture independently by adjusting the vibration intensity and frequency of the vibration motor. At the same time, the real-time pelvis movement data and correction execution situation are transmitted and stored to the user terminal device, and the user can intuitively view the history of the change of the pelvis posture, the correction record and the current posture state, etc. information through the APP.

[0081] Specifically, the hyperparameters of the preliminary improved SVM model are optimized, the optimal weight vector ω and the bias term b thereof are updated, and a trained improved SVM model is obtained, including:

[0082] randomly selecting multiple hyperparameters in the preliminary improved SVM model to form a first generation data set;

[0083] The fitness function is used to calculate the fitness of each individual in the first generation data set, and it is judged whether the individual meets the termination condition at this time, if it meets, the optimal weight vector ω and the bias term b in the preliminary improved SVM model are output and updated;

[0084] If it does not meet, the probability of each individual being selected in the next selection is calculated according to the obtained fitness and the probability formula of the individual i being selected.

[0085] According to the calculated probability, M individuals with large probability are selected to form the second generation data set; wherein, a certain data can be repeatedly selected, and M < N;

[0086] The data in the second generation data set is based on the mating probability pc, and the "crossing" operation is performed to generate N-M data, and the "mutation" operation is performed on the generated N-M data to obtain the N-M data set after mutation; then, the data in the second generation data set is generated in the form of "copying" to generate L data, and the L data is merged with the N-M data set after mutation to generate the third generation data set;

[0087] If the termination condition is met, the optimal weight vector ω and the bias term b in the preliminary improved SVM model are output and updated based on the third generation data set, the super parameter optimization of the preliminary improved SVM model is realized, and the trained improved SVM model is obtained.

[0088] Specifically, the performance of the final improved SVM prediction model is measured by calculating the correlation coefficient R and the relative error σ based on the mean absolute error and the root mean square error, and the performance of the final improved SVM prediction model is measured by the correlation coefficient R and the relative error σ.

[0089] In one specific embodiment of the present application, an intelligent wearable pelvis posture real-time correction device, as shown in the accompanying drawings, comprises: Figure 2

[0090] The intelligent waistband is made of adjustable elastic material, and the intelligent waistband is internally provided with a pelvis posture monitoring sensor, a microprocessor, a Bluetooth module and a battery; the two sides of the intelligent waistband are connected through a sticker.

[0091] The mobile terminal is connected with the intelligent waistband through Bluetooth, and is used for receiving and storing the data transmitted by the intelligent waistband.

[0092] ​A small vibration motor is installed in the inner ring of the intelligent waistband, and when an abnormal posture of the pelvis is detected, the vibration motor is controlled to generate vibrations of different frequencies and intensities according to different types of abnormalities, reminding the user to adjust the posture in time. The pelvis posture monitoring sensor device can measure the forward, backward, left and right tilts of the pelvis by being placed only on the position of the lumbar vertebra 5-sacrum;

[0093] Specifically, the pelvis posture monitoring sensor is used to collect the motion, angle and pressure data of the pelvis in real time and transmit them to the microprocessor for analysis and processing.

[0094] Specifically, the data collected by the pelvis posture monitoring sensor is fused by a Kalman filtering algorithm; the pelvis posture monitoring sensor includes an acceleration sensor, a gyroscope sensor and a pressure sensor; the algorithm can weight and fuse the data of different sensors inside according to the measurement noise and dynamic characteristics of the pelvis posture monitoring sensor, to obtain more accurate and stable pelvis motion state data. When the user performs complex movements, the acceleration sensor may be disturbed by the outside world, while the gyroscope sensor is relatively stable. The Kalman filtering algorithm automatically adjusts the weight according to the real-time noise estimation, and more data of the gyroscope sensor is collected to ensure that the final collected data can truly reflect the motion of the pelvis.

[0095] In order to ensure the synchronization of the data of the three sensors, a trigger mechanism based on clock synchronization is adopted. The internal clock of the microprocessor is used as the reference, and a synchronization signal is sent to the three sensors every 10 ms. The sensors collect and transmit data at the same time after receiving the synchronization signal, ensuring that the data of different sensors are aligned in time, and providing accurate time series data for subsequent analysis and processing.

[0096] At the same time, through the data of the acceleration sensor, the gyroscope sensor and the pressure sensor, when the user wears the waistband, due to the physiological structure characteristics of the lumbar vertebra 5-sacrum, different positions of the pressure sensor will sense different pressure values. According to the pressure distribution characteristics of the waistband, the correct position of the waistband is determined by analyzing and processing the pressure sensor data.

[0097] Specifically, the microprocessor collects pressure data from the pressure sensor at a certain sampling frequency and stores these data for subsequent analysis.

[0098] A pressure distribution characteristic model is established, and sample data is collected: a plurality of volunteers of different body types, different genders and different ages are invited to collect the pressure sensor data of their lumbar vertebra 5-sacrum parts under the condition of correctly wearing the intelligent waistband, to form a sample data set.

[0099] The sample data set is processed to extract features of the pressure distribution, such as the position of the pressure peak, the change in pressure gradient, the proportional relationship of pressure values in different regions, etc. Statistical analysis methods are used to calculate the mean, standard deviation, and other statistical quantities of each feature, thereby establishing a pressure distribution feature model under normal conditions.

[0100] The real-time collected pressure sensor data is preprocessed, and the same pressure distribution features as in the pressure distribution feature model are extracted from the preprocessed data.

[0101] A reasonable threshold range is set for each pressure distribution feature, and the real-time extracted pressure distribution feature values are compared with the threshold range. If all feature values are within the threshold range, it is considered that the waistband is in the correct position; otherwise, it is judged that the position is incorrect, and a position adjustment prompt is issued through the vibration motor.

[0102] When the waistband position is judged to be incorrect, a specific frequency and intensity of vibration is generated through the vibration motor (such as the two orange vibration motors near the pelvic posture monitoring sensor as shown in Figure 2 , or a sound is emitted through the mobile terminal APP to remind the user to adjust the waistband position.

[0103] During the user's adjustment of the waistband position, the pressure sensor data is continuously collected and analyzed until the waistband is judged to be in the correct position.

[0104] Specifically, in addition to allowing the user to adjust the waistband position, view the history of pelvic posture changes, correction records, and current posture status, etc., the mobile terminal APP also supports individualized setting functions. The user can customize the sensitivity of abnormal reminders according to their own physical condition, exercise habits, and working environment, etc. For example, for users who frequently engage in high-intensity physical labor, the threshold value of abnormal reminders can be appropriately increased to avoid frequent reminders due to normal physical movement; while for users who sit for long periods of time, the threshold value can be lowered to more promptly discover pelvic posture problems.

[0105] At the same time, the APP has built-in professional exercise rehabilitation courses and suggestion modules. According to the type and degree of the user's pelvic posture abnormalities, individualized exercise rehabilitation programs are provided for the user. For example, for users with a forward-leaning pelvis, some exercise courses that strengthen abdominal muscles and stretch hip flexor muscles are recommended, such as supine leg lift, hip stretching, etc., with detailed action demonstrations and explanation videos. At the same time, the exercise program is dynamically adjusted according to the user's correction progress, helping the user gradually restore normal pelvic posture.

[0106] Specifically, the adjustable elastic material adopted by the intelligent waistband has good air permeability, which can effectively reduce the stuffy feeling when the user wears it; at the same time, it has high wear resistance and can withstand friction and stretching in daily use, ensuring the service life of the product.

[0107] In a specific embodiment of the present application, the specific mapping relationship between different pelvic posture abnormality types and the vibration motor vibration intensity and frequency includes:

[0108] 1. Pelvic Anteversion

[0109] Mild Abnormality: When the pelvic anteversion is detected to be in a mild abnormality, that is, the anteversion angle is between 5°-10°, the vibration motor vibrates to remind the user at a frequency of 20Hz and an intensity of 50mN. This lower frequency and intensity vibration can gently make the user aware of the small problem of pelvic posture without being disturbed too much.

[0110] Moderate Abnormality: If the anteversion angle is between 10°-15°, it is a moderate abnormality, at this time the frequency of the vibration motor is increased to 40Hz and the intensity is increased to 100mN. The increase of frequency and intensity can more obviously attract the user's attention and remind him to adjust the posture as soon as possible.

[0111] Severe Abnormality: When the anteversion angle exceeds 15°, it is a severe abnormality, and the vibration motor vibrates at a frequency of 60Hz and an intensity of 150mN. Such strong vibration can make the user clearly perceive the serious abnormality of the pelvic posture in any activity state.

[0112] Pelvic Retroversion

[0113] Mild Abnormality: The retroversion angle between 3°-6° is a mild abnormality, and the vibration motor vibrates at a frequency of 18Hz and an intensity of 40mN. The retroversion angle is relatively small compared to the anteversion angle, so the initial frequency and intensity are also slightly lower.

[0114] Moderate Abnormality: When the retroversion angle is between 6°-9°, it is a moderate abnormality, the vibration frequency is increased to 35Hz, and the intensity is changed to 80mN.

[0115] Severe Abnormality: When the retroversion angle exceeds 9°, it is a severe abnormality, the vibration frequency reaches 55Hz, and the intensity is 120mN.

[0116] Pelvic Lateral Tilting

[0117] Mild Abnormality: The lateral tilting angle between 2°-4° is a mild abnormality, and the vibration motor vibrates at a frequency of 15Hz and an intensity of 30mN. Since lateral tilting may be less noticeable than anteversion and retroversion, it is also given some reminders when it is a mild abnormality.

[0118] Moderate abnormality: 4-6° of lateral inclination angle, 30Hz of vibration frequency, 60mN of intensity.

[0119] Severe abnormality: more than 6° of lateral inclination angle, 50Hz of vibration frequency, 100mN of intensity.

[0120] 2. Dynamic adjustment of reminder signal according to abnormality degree

[0121] When the abnormality degree increases, the vibration intensity and frequency increase linearly. Specifically, for every degree of increase in abnormality angle, the frequency increases at a certain slope and the intensity increases by a fixed increment in the mild-to-moderate and moderate-to-severe intervals of the corresponding abnormality type. For example, in the moderate abnormality interval of pelvic anteversion (10°-15°), for every 1° increase in anteversion angle, the frequency increases by 4Hz ((40-20) / (15-10)) and the intensity increases by 10mN ((100-50) / (15-10)).

[0122] At the same time, in order to avoid the user's adaptation to the reminder signal and ignore the reminder, when the abnormality persists for a certain period of time, the frequency is additionally increased by 5Hz and the intensity is additionally increased by 10mN every 3 minutes based on the current abnormality degree corresponding to the basic frequency and intensity, until the user adjusts the posture to reduce or eliminate the abnormality degree.

[0123] 3. Duration and interval time setting of reminder signal

[0124] The duration of the reminder signal is 3 seconds each time and the interval time is 15 seconds. Such a setting can ensure that the user has enough time to perceive the vibration reminder and will not be bored by too frequent reminders. When the abnormality degree increases, the duration and interval time remain unchanged, but the vibration intensity and frequency will increase according to the dynamic adjustment rules described above. If the user still does not adjust the posture to reduce the abnormality after 3 consecutive reminders, the interval time is shortened to 10 seconds to urge the user to adjust the pelvic posture more intensively. When the abnormality degree decreases or is eliminated, the initial duration and interval time settings are restored.

[0125] The embodiments of the present application can be applied in gyms, sports training institutions and other places. Coaches can monitor the pelvic posture of athletes or fitness enthusiasts in real time through the system, timely find possible incorrect postures during exercise, avoid exercise injuries caused by long-term incorrect postures, and develop more scientific training plans for users according to the monitoring data to improve training effectiveness.

[0126] It can also be used as an auxiliary tool for the rehabilitation treatment of patients with pelvic injury or disease. Doctors can real-time understand the recovery of the patient's pelvic posture through the system and adjust the rehabilitation treatment plan, such as the intensity and frequency of physical therapy, the action and difficulty of rehabilitation training, etc., to help the patient recover faster.

[0127] In nursing homes or family care, using the device can real-time monitor the pelvic posture of the elderly, and timely remind the nursing staff or family members when there is an abnormality, and take corresponding measures, such as helping the elderly adjust the posture, arranging further examination and treatment, etc., to improve the life quality and safety of the elderly.

[0128] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts are referred to the method part.

[0129] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A real-time correction method for the posture of a pelvis, characterized in that, The application relates to a method for monitoring the posture of a pelvis, and a device for monitoring the posture of a pelvis. The application comprises the following steps: Integrating a pelvis posture monitoring sensor in a smart wearable device to collect real-time pelvis movement data of a user; Processing and analyzing the pelvis movement data to extract a set of feature parameters related to the posture of the pelvis; Comparing the extracted set of feature parameters with pre-stored normal pelvis postures for analysis; Using machine learning algorithms and pattern recognition techniques to determine whether the current pelvis posture is within the normal range; If an abnormal pelvis posture is detected, determining the type and degree of the abnormality; Based on the determination result, sending different posture abnormality warning signals to the user; The set of feature parameters related to the posture of the pelvis is extracted, including: Based on a feature pyramid network, an improved side fusion network is introduced to establish a pelvis posture feature extraction network; The set of feature parameters related to the posture of the pelvis is extracted through the pelvis posture feature extraction network; Based on a feature pyramid network, an improved side fusion network is introduced to establish a pelvis posture feature extraction network, including: The pelvis movement data is processed in stages using a feature pyramid network; in each stage, a convolution operation is performed, followed by a pooling operation, to obtain feature parameters {C1, C2, C3, C4, C5} at different stages; Using a top-down fusion path, the feature parameters {C5} at the top layer are enlarged using an up-sampling method to make their dimension size consistent with the feature parameters at the previous stage; The enlarged feature parameters are added to the corresponding elements of the feature parameters at the previous stage to obtain fused feature parameters {P2, P3, P4, P5}; The side fusion network structure is introduced, and in the same stage and at the same scale, {N1} is directly equal to {C1}, and {N5} is directly equal to {C5}, which remain unchanged; On this basis, a jump connection path is added from {C2, C3, C4} to {N2, N3, N4}; through the jump connection path, the feature parameters of {C2, C3, C4} are side-fused with the feature parameters {P2, P3, P4} after top-down fusion to generate a set of feature parameters {N1, N2, N3, N4, N5} containing the feature information of the current layer and the fused feature information of the top layer; Using machine learning algorithms and pattern recognition techniques to determine whether the current pelvis posture is within the normal range, including: Selecting a feature parameter in the set of feature parameters to construct a new optimization problem, and replacing the basic optimization problem in the traditional SVM model with the new optimization problem; Selecting a sigmoid kernel function as the kernel function of the traditional SVM model to obtain a preliminary improved SVM model; Optimizing the hyperparameters of the preliminary improved SVM model to update the optimal weight vector omega and the bias term b, and obtaining the trained improved SVM model; Based on the mean absolute error and the root mean square error, the values of the correlation coefficient R and the relative error sigma are obtained, and R and sigma are used to evaluate the performance of the trained improved SVM model; meanwhile, the complexity of the trained improved SVM model is controlled by adjusting the regularization parameter C; The improved SVM model after evaluation is used for new data in the feature parameter set, and ω · x in the improved SVM model after evaluation is calculated i + b, the data is assigned to the category with a calculated value greater than 0.05, and an improved SVM model after classification is obtained; The other feature parameters in the feature parameter set are repeated with the above operation, and the weight and bias of the preliminary improved SVM model are output and updated to complete the preliminary improved SVM model learning, and a final improved SVM prediction model is obtained. 2.The intelligent wearable pelvis posture real-time correction method according to claim 1, characterized in that, Also include: Real-time pelvic motion data and correction execution are transmitted and stored to the user terminal device. 3.The intelligent wearable pelvis posture real-time correction method according to claim 1, characterized in that, The preliminary improved SVM model is super parameter optimized, and the optimal weight vector ω and bias term b are updated to obtain the trained improved SVM model, including: Randomly selecting multiple super parameters in the preliminary improved SVM model to form a first generation data set; The fitness function is used to calculate the fitness of each individual in the first generation data set, and it is judged whether the individual meets the termination condition at this time, if it meets, the optimal weight vector ω and bias term b of the preliminary improved SVM model are output and updated; If not, the probability of each individual being selected in the next selection is calculated according to the obtained fitness and the probability formula of individual i being selected; According to the calculated probability, M individuals with large probability are selected to form a second generation data set; wherein, a certain data can be repeatedly selected, and M < N; The data in the second generation data set is operated based on the mating probability pc to generate N-M data, and the N-M data is subjected to mutation operation to obtain a N-M data set; then, L data are generated in the form of "copy" from the second generation data set, and the L data are merged with the N-M data set to generate a third generation data set; If the termination condition is met, the optimal weight vector ω and bias term b of the preliminary improved SVM model are output and updated based on the third generation data set to realize the super parameter optimization of the preliminary improved SVM model, and the trained improved SVM model is obtained. 4.The intelligent wearable pelvis posture real-time correction method according to claim 3, characterized in that, The performance of the final improved SVM prediction model is measured by calculating the correlation coefficient R and the relative error σ based on the mean absolute error and the root mean square error, and the performance of the final improved SVM prediction model is measured by the correlation coefficient R and the relative error σ.

5. A smart wearable pelvis posture real-time correction device, applying the smart wearable pelvis posture real-time correction method of any one of claims 1-4, characterized in that, Including: The intelligent waistband is made of elastic material, and the intelligent waistband is internally provided with a pelvic posture monitoring sensor, a microprocessor, a Bluetooth module and a battery. The mobile terminal is connected with the intelligent waistband through Bluetooth, and is used for receiving and storing the data transmitted by the intelligent waistband. A vibration motor is installed in the inner ring of the intelligent waistband, and when the pelvic posture is detected to be abnormal, the vibration motor is controlled to generate vibrations with different frequencies and intensities according to different types of abnormalities, reminding the user to adjust the posture in time. 6.The intelligent wear-type real-time pelvis posture correction device according to claim 5, characterized in that, The pelvic posture monitoring sensor is used to collect the motion, angle and pressure data of the pelvis in real time, and transmit them to the microprocessor for analysis and processing.

Images