A sitting posture classification method based on a seat cushion

Abstract

A sitting posture classification method based on a seat cushion, which belongs to the technical field of sitting posture classification. The present invention solves the problems of low classification accuracy, high cost and poor adaptability existing in the traditional sitting posture classification method. In the present invention, force-sensitive resistors and accelerometers corresponding to the ischial tuberosities and thigh regions of the human body are fixed on the seat cushion body, which can comprehensively and effectively collect human sitting posture signal data and effectively reduce the cost of sitting posture classification. By solving the pressure-acceleration sitting posture signal data correlation degree index matrix, the data strongly related to sitting posture classification are screened out; the signal data strongly related to sitting posture classification screened out are fused to effectively establish the connection between the data. Finally, by constructing the correlation degree feature distance and the correlation degree feature distance classification algorithm, the signal data to be classified and the fusion result are processed together to obtain the human sitting posture classification result. The method of the present invention can be applied to the field of human sitting posture classification.

Description

Technical Field

[0001] The present invention belongs to the technical field of sitting posture classification, and particularly relates to a sitting posture classification method based on a seat cushion. Background Art

[0002] In modern society, people are in a sitting posture for a long time during work, study and life. However, incorrect sitting postures may lead to various health problems, such as scoliosis, cervical and lumbar spine diseases, etc., and also affect work efficiency and quality of life. Therefore, accurately detecting and classifying sitting postures is of great significance for timely reminding people to maintain correct sitting postures and preventing health problems.

[0003] Current sitting posture detection methods and technologies are diverse. For example, Literature 1 (Roland Z, Matteo T, Plüss Stefan, et al. Application of Machine Learning Approaches for Classifying Sitting Posture Based on Force and Acceleration Sensors. BioMed Research International, 2016: 1-9. DOI: 10.1155 / 2016 / 5978489.) uses 16 pressure sensors and 1 inertial sensor. The pressure sensors are respectively placed in the seat pan, on the backrest and on each armrest, and the inertial sensor is placed behind the backrest. However, these sensors are just simply listed, and there is basically no connection between the sensors. Moreover, for the sensors arranged on the backrest and armrests, they may not cause data changes in these sensors for different people. Therefore, the error of sensor data measurement is relatively large. At the same time, due to the lack of connection between the data of each sensor, the accuracy of sitting posture classification based on the data collected by the sensors is relatively low, and using a large number of pressure sensors will lead to a relatively high overall cost. Literature 2 (Kerr J, Carlson J, Godbole S, et al. Improving Hip-Worn Accelerometer Estimates of Sitting Using Machine Learning Methods. Medicine&Science in Sports&Exercise, 2018: 1. DOI: 10.1249 / MSS.0000000000001578.) only involves accelerometers. Because the collected data is often not comprehensive and accurate enough, it is difficult to accurately distinguish different types of sitting postures. And the accelerometer is worn in front of the subject's thigh, which belongs to an invasive measurement of the human body.Both Document 3 (Ran X, Wang C, Xiao Y, et al. A portable sitting posture monitoring system based on a pressure sensor array and machine learning[J]. Sensors and Actuators A: Physical, 2021, 331: 112900-. DOI: 10.1016 / j.sna.2021.112900.) and Document 4 (Jawad A, Johan, Sidén, Henrik A. A Proposal of Implementation of Sitting Posture Monitoring System for Wheelchair Utilizing Machine Learning Methods. Sensors. 2021 21(19): 6349. DOI: 10.3390 / s21196349.) use a force sensor array, which leads to a significant increase in system cost and the accuracy of sitting posture classification is still relatively limited. In addition, some of the existing technologies perform poorly in terms of adaptability, unable to effectively reflect changes in sitting postures and difficult to meet the needs of different people.

[0004] In summary, the traditional sitting posture classification methods still have problems such as low classification accuracy, high cost and poor adaptability. Therefore, in order to overcome the deficiencies of the existing technologies, it is an urgent problem to provide a more advanced and effective solution for sitting posture detection and classification. Summary of the Invention

[0005] The object of the present invention is to solve the problems of low classification accuracy, high cost and poor adaptability existing in the traditional sitting posture classification methods, and a sitting posture classification method based on a cushion is proposed.

[0006] The technical solution adopted by the present invention to solve the above technical problems is: a sitting posture classification method based on a cushion, and the method specifically includes the following steps:

[0007] Step 1: Obtain a reference data matrix for sitting posture classification from the server side;

[0008] Step 2: When the user sits on the cushion, use each force-sensitive resistor and accelerometer on the cushion to collect the user's signal data, and then select the data associated with sitting posture classification from the collected user signal data;

[0009] Step 3: Obtain the sitting posture classification result of the user according to the data selected in Step 2 and the reference data matrix.

[0010] The beneficial effects of the present invention are as follows:

[0011] In the present invention, 4 force - sensitive resistors and 2 accelerometers are fixed on the seat cushion body, and these sensors respectively correspond to the ischial tuberosities and thigh regions of the human body, so as to comprehensively and effectively collect human sitting posture signal data, providing a reliable data source for accurately classifying sitting postures. By solving the correlation degree index matrix of pressure - acceleration sitting posture signal data, the strong correlation degree index data of pressure - acceleration sitting posture signals with the absolute value of the correlation degree index above 0.8 are screened out; by fusing the screened signal data with strong correlation degree indexes for sitting posture classification, the connection between data is effectively established. Finally, by constructing the correlation degree feature distance and the correlation degree feature distance classification algorithm, the signal data to be classified and the fusion result are processed together, thus effectively improving the accuracy and efficiency of classifying different human sitting postures. While effectively distinguishing a variety of common sitting postures, the adaptability of the classification method is improved.

[0012] At the same time, since the method of the present invention does not require a large number of force sensors, the cost of sitting posture classification is effectively reduced. Brief Description of the Drawings

[0013] Figure 1 is a flowchart of a sitting posture classification method based on a seat cushion according to the present invention;

[0014] Figure 2 is a schematic structural diagram of a seat cushion for sitting posture classification according to the present invention;

[0015] Among them, seat cushion body 1, first force - sensitive resistor 2, second force - sensitive resistor 3, third force - sensitive resistor 4, fourth force - sensitive resistor 5, first accelerometer 6, second accelerometer 7, first fixing strap 8, second fixing strap 9, third fixing strap 10, fourth fixing strap 11, first central axis 12, second central axis 13;

[0016] Figure 3 is a numerical normalization curve graph of the force - sensitive resistor collecting the signal data of the subject;

[0017] Figure 4 is a numerical normalization curve graph of the accelerometer collecting the signal data of the subject;

[0018] Figure 5 is the maximum correlation degree index under various sitting postures of the subject;

[0019] Figure 6 is a schematic diagram of a confusion matrix for classifying and predicting the subject data by using the correlation degree feature distance classification algorithm. Detailed Embodiments

[0020] Detailed Embodiment 1: In combination withFigure 2 Describe this embodiment. A cushion for sitting posture classification according to this embodiment, the cushion includes a cushion main body 1, a first force-sensitive resistor 2, a second force-sensitive resistor 3, a third force-sensitive resistor 4, a fourth force-sensitive resistor 5, a first accelerometer 6, a second accelerometer 7, a first central axis 12 and a second central axis 13; wherein:

[0021] The first force-sensitive resistor 2, the second force-sensitive resistor 3, the third force-sensitive resistor 4, the fourth force-sensitive resistor 5, the first accelerometer 6 and the second accelerometer 7 are all placed on the surface of the cushion main body 1;

[0022] The first force-sensitive resistor 2 and the second force-sensitive resistor 3 are symmetrically arranged on both sides of the first central axis 12 of the cushion main body 1;

[0023] The third force-sensitive resistor 4 and the fourth force-sensitive resistor 5 are symmetrically arranged on both sides of the first central axis 12 of the cushion main body 1;

[0024] The first accelerometer 6 and the second accelerometer 7 are symmetrically arranged on both sides of the first central axis 12 of the cushion main body 1. The first accelerometer 6 is placed between the first force-sensitive resistor 2 and the third force-sensitive resistor 4, and the second accelerometer 7 is placed between the second force-sensitive resistor 3 and the fourth force-sensitive resistor 5, and the first accelerometer 6 and the second accelerometer 7 are placed on the second central axis 13 of the cushion main body 1.

[0025] Specific embodiment two: The difference between this embodiment and the first specific embodiment is that the shape of the cushion main body 1 is rectangular.

[0026] Other steps and parameters are the same as those in the first specific embodiment.

[0027] The first force-sensitive resistor 2 and the second force-sensitive resistor 3 respectively correspond to the left and right ischial tuberosity areas of the user, and the third force-sensitive resistor 4 and the fourth force-sensitive resistor 5 respectively correspond to the left and right thigh areas of the user. The length of the cushion main body 1 is set to 55 cm and the width is set to 45 cm. Among them, the first force-sensitive resistor 2 and the second force-sensitive resistor 3 are 11 cm apart, the third force-sensitive resistor 4 and the fourth force-sensitive resistor 5 are 47.5 cm apart. While the first accelerometer 6 and the second accelerometer 7 are symmetrically arranged on both sides of the first central axis 12, they are also located on the second central axis 13, and the first accelerometer 6 and the second accelerometer 7 are 24 cm apart. The parameter values in the present invention can be taken as the above values, but are not limited to the above values. For example, different parameter designs can be carried out for children and adults respectively.

[0028] Embodiment 3: The difference between this embodiment and Embodiment 1 or 2 is that the cushion further includes a first fixing strap 8, a second fixing strap 9, a third fixing strap 10 and a fourth fixing strap 11, and the first fixing strap 8, the second fixing strap 9, the third fixing strap 10 and the fourth fixing strap 11 are used to fix the cushion body 1 on the seat surface.

[0029] Other steps and parameters are the same as those in Embodiment 1 or 2.

[0030] The third fixing strap 10 and the fourth fixing strap 11 are respectively arranged on two sides of the cushion body 1 parallel to the first central axis (corresponding to the wide sides of the rectangular cushion in the present invention). Among the two sides of the cushion body 1 parallel to the second central axis, the first fixing strap 8 and the second fixing strap 9 are arranged on the side close to the back. The first fixing strap 8, the second fixing strap 9, the third fixing strap 10 and the fourth fixing strap 11 are used to cooperate with each other to fix the cushion body 1 on the seat surface.

[0031] Embodiment 4: Combine Figure 1 to illustrate this embodiment. A sitting posture classification method based on a cushion according to this embodiment specifically includes the following steps:

[0032] Step 1: Obtain a reference data matrix for sitting posture classification from the server side;

[0033] Step 2: When the user sits on the cushion, use the force-sensitive resistors and accelerometers on the cushion to collect the user's signal data, and then select the data associated with sitting posture classification from the collected user signal data (the data associated with coordinate classification here is the data determined when calculating the reference data matrix and associated with sitting posture classification);

[0034] Step 3: Obtain the sitting posture classification result of the user according to the data selected in Step 2 and the reference data matrix.

[0035] Embodiment 5: The difference between this embodiment and Embodiment 4 is that in Step 1, the generation method of the reference data matrix stored on the server side is as follows:

[0036] Step 11: A total of R subjects are selected. Each subject sits on the cushion in various sitting postures. That is, for the selected r-th subject, P groups of signal data are collected in each sitting posture of the r-th subject, and each group of signal data includes force-sensitive resistor signal data and accelerometer signal data, r = 1, 2,..., R;

[0037] And R×P groups of signal data when there is no subject on the cushion are collected as baseline signals;

[0038] Steps 1 and 2: Numerically standardize the signal data of each subject. Taking the r-th subject as an example:

[0039] Numerically standardize the data collected by the first force-sensitive resistor in each group of signal data corresponding to the r-th subject;

[0040] Numerically standardize the data collected by the second force-sensitive resistor in each group of signal data corresponding to the r-th subject;

[0041] Numerically standardize the data collected by the third force-sensitive resistor in each group of signal data corresponding to the r-th subject;

[0042] Numerically standardize the data collected by the fourth force-sensitive resistor in each group of signal data corresponding to the r-th subject;

[0043] Numerically standardize the acceleration data in the x-axis direction collected by the first accelerometer in each group of signal data corresponding to the r-th subject;

[0044] Numerically standardize the acceleration data in the y-axis direction collected by the first accelerometer in each group of signal data corresponding to the r-th subject;

[0045] Numerically standardize the acceleration data in the z-axis direction collected by the first accelerometer in each group of signal data corresponding to the r-th subject;

[0046] Numerically standardize the acceleration data in the x-axis direction collected by the second accelerometer in each group of signal data corresponding to the r-th subject;

[0047] Numerically standardize the acceleration data in the y-axis direction collected by the second accelerometer in each group of signal data corresponding to the r-th subject;

[0048] Numerically standardize the acceleration data in the z-axis direction collected by the second accelerometer in each group of signal data corresponding to the r-th subject;

[0049] Then, record the result of numerically standardizing the i-th group of signal data corresponding to the r-th subject as where, is the numerically standardized value of the data collected by the first force-sensitive resistor in the i-th group of signal data, is the numerically standardized value of the data collected by the second force-sensitive resistor in the i-th group of signal data, is the numerically standardized value of the data collected by the third force-sensitive resistor in the i-th group of signal data, is the numerically standardized value of the data collected by the fourth force-sensitive resistor in the i-th group of signal data, is the numerical normalization value of the acceleration data in the x-axis direction collected by the first accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the y-axis direction collected by the first accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the z-axis direction collected by the first accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the x-axis direction collected by the second accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the y-axis direction collected by the second accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the z-axis direction collected by the second accelerometer in the i-th group of signal data;

[0050] The numerical normalization result of the signal data collected by the force-sensitive resistor is as Figure 3 shown, and the numerical normalization result of the signal data collected by the accelerometer is as Figure 4 shown; similarly, the data of each subject is numerically normalized;

[0051] Step Thirteen: Using the numerically normalized data corresponding to any one subject, determine the data whose association degree index with sitting posture classification is greater than the threshold, and use the determined data as the data related to sitting posture classification in each group;

[0052] Step Fourteen: Retain the data related to sitting posture classification in all groups, and then fuse the retained force-sensitive resistor data and accelerometer data in each group respectively (here, the dynamic filtering method is used for data fusion) to obtain the fused data of each group;

[0053] Take the fused data of each group as a row of the reference data matrix respectively, and use the fused data of each group to form the reference data matrix.

[0054] Other steps and parameters are the same as those in the fourth specific implementation manner.

[0055] Specific Implementation Manner Six: The difference between this implementation manner and the fourth or fifth specific implementation manner is that the sitting postures of the subjects specifically include sitting upright, leaning left, leaning right, leaning forward, leaning backward, shaking the left leg, shaking the right leg, crossing the left leg, and crossing the right leg.

[0056] Other steps and parameters are the same as those in the fourth or fifth specific implementation manner.

[0057] The characteristics of sitting upright are: the spine is in a neutral position, the head maintains natural balance, without obvious tilt or rotation.

[0058] The characteristics of leaning left are: the body center of gravity shifts to the left, the left lumbar muscles contract, and the right muscles stretch.

[0059] The right-leaning feature is that the center of body gravity shifts to the right, the muscles on the right side of the waist contract, and the muscles on the left side are stretched.

[0060] The forward-leaning feature is that the center of body gravity moves forward, the back muscles are stretched, and the abdominal muscles contract.

[0061] The backward-leaning feature is that the center of body gravity moves backward, the abdominal muscles are stretched, and the back muscles contract.

[0062] The left leg shaking feature is that the muscles of the left leg contract and relax rapidly, producing a shaking effect, while the right leg and the upper body remain stable.

[0063] The right leg shaking feature is that the muscles of the right leg contract and relax rapidly, producing a shaking effect, while the left leg and the upper body remain stable.

[0064] The left crossed-leg feature is that the left leg abducts, the right leg adducts, and the center of body gravity slightly shifts to the right.

[0065] The right crossed-leg feature is that the right leg abducts, the left leg adducts, and the center of body gravity slightly shifts to the left.

[0066] Specific Embodiment 7: The difference between this embodiment and one of Embodiments 4 to 6 is that the specific process of Step 13 is as follows:

[0067] For any sitting posture, the correlation index between the data collected by the first force-sensitive resistor and the acceleration data in the x-axis direction collected by the first accelerometer for the r-th subject in this sitting posture is:

[0068]

[0069] where

[0070] Then the correlation index matrix Q corresponding to the r-th subject in this sitting posture is:

[0071]

[0072] Calculate the absolute value of each element in the matrix Q respectively, obtain all the elements in the matrix Q whose absolute value is greater than 0.8, and then determine the data associated with this sitting posture according to the obtained all elements, that is, if the absolute value of r a,b is greater than 0.8, then both a and b are the data associated with this sitting posture,

[0073] Similarly, determine the data associated with each sitting posture of the r-th subject respectively, and use the determined all data as the data associated with sitting posture classification.

[0074] Other steps and parameters are the same as those in one of Embodiments 4 to 6.

[0075] Embodiment 8: The difference between this embodiment and one of Embodiments 4 to 7 is that in Step 14, the combined data of each group is used to form a reference data matrix, specifically:

[0076]

[0077] where n is the total number of groups of data collected in Step 11, M is the reference data matrix, G is the number of data types associated with sitting posture classification, g = 1, 2,..., G, and (x i1 , x i2 , …, x iG ) is the i-th group of combined data.

[0078] Other steps and parameters are the same as those in one of Embodiments 4 to 7.

[0079] Since the data determined to be associated with sitting posture classification in the embodiments of the present invention is Therefore, the value of G is 8 at this time.

[0080] Embodiment 9: The difference between this embodiment and one of Embodiments 4 to 8 is that the specific process of Step 3 is as follows:

[0081] Step 31: Denote the data selected from the collected user signal data and associated with sitting posture classification as

[0082]

[0083] Step 32: Calculate the correlation degree characteristic distance between and each row vector in the reference data matrix M;

[0084]

[0085] where x i is the i-th row vector in the reference data matrix M, x i = (x i1 , x i2 , …, x iG ), T represents the transpose operation, S is the covariance matrix, S -1 is the inverse of the covariance matrix, is the correlation degree characteristic distance between and the i-th row vector in the reference data matrix M;

[0086] Step 33: Obtain the sitting posture classification result according to :

[0087]

[0088] where C is the sitting posture classification set, is the classification result of the user's sitting posture, I(y i = c) is an indicator function, y i is the true category of the i-th row vector in the reference data matrix M. When y i = c, the value of I(y i = c) is 1. When y i ≠ c, the value of I(y i = c) is 0.

[0089] Other steps and parameters are the same as those in one of the fourth to eighth specific embodiments.

[0090] Next, the steps three three in this embodiment will be further described in detail. Each category c in the sitting posture classification set C is substituted into the formula for calculation. For example, when the category c is left tilt, if the true category of the i-th row vector is left tilt, then the value of I(y i = c) is 1, otherwise the value of I(y i = c) is 0. Then, combined with the previously calculated the value corresponding to the current category c can be obtained. After traversing all the categories in the set C, by comparing the values corresponding to each category the maximum value is selected, and the category corresponding to the maximum value is used as the classification result of the user's sitting posture. value is used as the classification result of the user's sitting posture.

[0091] Specific embodiment ten: The difference between this embodiment and one of the fourth to ninth specific embodiments is that the covariance matrix S is:

[0092]

[0093] where σ jk is an element in the covariance matrix, j = 1, 2,..., G, k = 1, 2,..., G;

[0094]

[0095] where

[0096] Other steps and parameters are the same as those in one of the fourth to ninth specific embodiments.

[0097] Example

[0098] Next, a sitting posture classification method based on a cushion of the present invention will be further described in detail with reference to the accompanying drawings. The sitting posture classification method specifically includes the following steps:

[0099] The present invention stores a reference data matrix for sitting posture classification on the server side for subsequent retrieval by different users, and the method for generating the reference data matrix is as follows:

[0100] First, 30 subjects are selected, including 15 males and 15 females. Each subject includes no person (baseline data) + 9 sitting posture data. For each sitting posture of each subject, 200 groups of stable output data are respectively intercepted. The data acquisition frequency is 10 Hz, and each group includes 10 data points (6 columns of data from 2 accelerometers and 4 columns of data from 4 force-sensitive resistors), that is, the present invention collects a total of n = 2000×30 groups of data. The collected data is processed as follows:

[0101] Step 1: A total of R subjects are selected. Each subject sits on the cushion in various sitting postures. That is, for the selected r-th subject, P groups of signal data are collected for each sitting posture of the r-th subject, and each group of signal data includes force-sensitive resistor signal data and accelerometer signal data, r = 1, 2,..., R;

[0102] And R×P groups of signal data when there is no subject on the cushion are collected as baseline signals;

[0103] Step 2: Numerically standardize the signal data of each subject respectively. Taking the r-th subject as an example:

[0104] Numerically standardize the data collected by the first force-sensitive resistor in each group of signal data corresponding to the r-th subject;

[0105] Numerically standardize the data collected by the second force-sensitive resistor in each group of signal data corresponding to the r-th subject;

[0106] Numerically standardize the data collected by the third force-sensitive resistor in each group of signal data corresponding to the r-th subject;

[0107] Numerically standardize the data collected by the fourth force-sensitive resistor in each group of signal data corresponding to the r-th subject;

[0108] Numerically standardize the x-axis direction acceleration data collected by the first accelerometer in each group of signal data corresponding to the r-th subject;

[0109] Numerically standardize the y-axis direction acceleration data collected by the first accelerometer in each group of signal data corresponding to the r-th subject;

[0110] Numerically standardize the z-axis direction acceleration data collected by the first accelerometer in each group of signal data corresponding to the r-th subject;

[0111] Perform numerical normalization on the acceleration data in the x-axis direction collected by the second accelerometer in each group of signal data corresponding to the r-th subject;

[0112] Perform numerical normalization on the acceleration data in the y-axis direction collected by the second accelerometer in each group of signal data corresponding to the r-th subject;

[0113] Perform numerical normalization on the acceleration data in the z-axis direction collected by the second accelerometer in each group of signal data corresponding to the r-th subject;

[0114] Then, record the numerical normalization result of the i-th group of signal data corresponding to the r-th subject as where, is the numerical normalization value of the data collected by the first force-sensitive resistor in the i-th group of signal data, is the numerical normalization value of the data collected by the second force-sensitive resistor in the i-th group of signal data, is the numerical normalization value of the data collected by the third force-sensitive resistor in the i-th group of signal data, is the numerical normalization value of the data collected by the fourth force-sensitive resistor in the i-th group of signal data, is the numerical normalization value of the acceleration data in the x-axis direction collected by the first accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the y-axis direction collected by the first accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the z-axis direction collected by the first accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the x-axis direction collected by the second accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the y-axis direction collected by the second accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the z-axis direction collected by the second accelerometer in the i-th group of signal data;

[0115] The numerical normalization results of the signal data collected by the force-sensitive resistors are as Figure 3 shown, and the numerical normalization results of the signal data collected by the accelerometers are as Figure 4 shown; Similarly, perform numerical normalization on the data of each subject;

[0116] Step 13: Use the numerically normalized data corresponding to any subject to determine the data whose correlation index with sitting posture classification is greater than the threshold, and use the determined data as the data related to sitting posture classification in each group; Specifically:

[0117] For any sitting posture, calculate the correlation index for this sitting posture using the data collected by the first force-sensitive resistor and the acceleration data in the x-axis direction collected by the first accelerometer for the r-th subject. It is:

[0118]

[0119] Where,

[0120] Then the correlation index matrix Q corresponding to the r-th subject in this sitting posture is:

[0121]

[0122] Calculate the absolute value of each element in matrix Q respectively, obtain all the elements in matrix Q whose absolute value is greater than 0.8, and then determine the data associated with this sitting posture according to the obtained all elements, that is, if the absolute value of r a,b is greater than 0.8, then both a and b are data associated with this sitting posture.

[0123] Similarly, determine the data associated with each sitting posture of the r-th subject respectively, and use the determined all data as the data associated with sitting posture classification.

[0124] It should be noted that as long as the absolute value of any element corresponding to data a in the correlation index matrix of any sitting posture is greater than 0.8, then data a is the data associated with sitting posture classification. And use the data associated with the sitting posture classification of the r-th subject as the data associated with the sitting posture classification of each subject, that is, it is considered that for each subject, the types of data associated with sitting posture classification are the same

[0125] The maximum correlation index of the subject in various sitting postures is as Figure 5 shown. In the present invention, the data determined to be associated with sitting posture classification by calculation is

[0126] Step 14. Retain the data associated with sitting posture classification in all groups, and then fuse the force-sensitive resistor data and accelerometer data retained in each group respectively (here, a dynamic filtering method is used for data fusion) to obtain the fused data of each group;

[0127] Take the fused data of each group as a row of the reference data matrix respectively, and use the fused data of each group to form the reference data matrix:

[0128]

[0129] where n is the total number of data groups collected in Step 1, M is the reference data matrix, g = 1, 2, …, 8, and (x i1 , x i2 , …, x i8 ) is the i-th group of data after fusion.

[0130] When the user sits on the seat cushion, various force-sensitive resistors and accelerometers on the seat cushion are used to collect the user's signal data, and then the data associated with the sitting posture classification is selected from the collected user signal data (the data associated with the coordinate classification here is the data associated with the sitting posture classification determined when calculating the reference data matrix);

[0131] And the sitting posture classification result of the user is obtained based on the data selected in Step 2 and the reference data matrix; the specific process is as follows:

[0132] The data associated with the sitting posture classification selected from the collected user signal data is denoted as

[0133]

[0134] Calculate respectively the correlation degree feature distance with each row vector in the reference data matrix M;

[0135]

[0136] where x i is the i-th row vector in the reference data matrix M, x i = (x i1 , x i2 , …, x i8 ), T represents the transpose operation, S is the covariance matrix, S -1 is the inverse of the covariance matrix, is the correlation degree feature distance with the i-th row vector in the reference data matrix M;

[0137] According to obtain the sitting posture classification result:

[0138]

[0139] where C is the sitting posture classification set, is the sitting posture classification result of the user, I(y i = c) is the indicator function, y i is the true category of the i-th row vector in the reference data matrix M, when y i = c, the value of I(y i = c) is 1, when y i ≠ c, I(y iThe value of = c) is 0.

[0140] In the present invention, a total of n = 60,000 groups of data are collected from the subjects. After initializing the number of data groups n' = 1, all the other groups of data except the n'-th group of data are used as reference data. The sitting posture represented by the n'-th group of data is predicted based on the reference data, and then the number of data groups n' = n' + 1 is set. The iteration stops until the number of data groups n' = 60,000, that is, the sitting posture prediction results for each group of data of the subjects are obtained respectively. The classification results of the subject data are as Figure 6 shown. Labels 0 to 9 represent no person, sitting upright, leaning left, leaning right, leaning forward, leaning backward, shaking the left leg, shaking the right leg, crossing the left leg, and crossing the right leg in sequence. TPR is the true positive rate, quantifying the ability of the quantization model to correctly identify positives in all actual positive cases. FNR is the false negative rate, reflecting the proportion of actual positive cases that the model fails to correctly classify as positive.

[0141] The above numerical examples of the present invention are only for illustrating in detail the calculation model and calculation process of the present invention, rather than limiting the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or variations can be made based on the above description. It is impossible to enumerate all the implementation manners here. Any obvious changes or variations derived from the technical solutions of the present invention still fall within the protection scope of the present invention.

Claims

1. A sitting posture classification method based on a seat cushion, characterized in that, The method specifically includes the following steps: Step 1: Obtain a reference data matrix for sitting posture classification from the server side; The generation method of the reference data matrix is as follows: Step 1-1: A total of R subjects are selected. Each subject sits on the cushion in various sitting postures. That is, for the selected r-th subject, P groups of signal data are collected under each sitting posture of the r-th subject, and each group of signal data includes force-sensitive resistor signal data and accelerometer signal data, where r = 1, 2, …, R; And R×P groups of signal data when there is no subject on the cushion are collected as baseline signals; Step 1-2: Numerically standardize the signal data of each subject. Taking the r-th subject as an example: Numerically standardize the data collected by the first force-sensitive resistor in each group of signal data corresponding to the r-th subject; Numerically standardize the data collected by the second force-sensitive resistor in each group of signal data corresponding to the r-th subject; Numerically standardize the data collected by the third force-sensitive resistor in each group of signal data corresponding to the r-th subject; Numerically standardize the data collected by the fourth force-sensitive resistor in each group of signal data corresponding to the r-th subject; Numerically standardize the x-axis direction acceleration data collected by the first accelerometer in each group of signal data corresponding to the r-th subject; Numerically standardize the y-axis direction acceleration data collected by the first accelerometer in each group of signal data corresponding to the r-th subject; Numerically standardize the z-axis direction acceleration data collected by the first accelerometer in each group of signal data corresponding to the r-th subject; Numerically standardize the x-axis direction acceleration data collected by the second accelerometer in each group of signal data corresponding to the r-th subject; Numerically standardize the y-axis direction acceleration data collected by the second accelerometer in each group of signal data corresponding to the r-th subject; Numerically standardize the z-axis direction acceleration data collected by the second accelerometer in each group of signal data corresponding to the r-th subject; The numerical normalization result of the i-th group of signal data corresponding to the r-th subject is denoted as where is the numerical normalization value of the data collected by the first force-sensitive resistor in the i-th group of signal data, is the numerical normalization value of the data collected by the second force-sensitive resistor in the i-th group of signal data, is the numerical normalization value of the data collected by the third force-sensitive resistor in the i-th group of signal data, is the numerical normalization value of the data collected by the fourth force-sensitive resistor in the i-th group of signal data, is the numerical normalization value of the acceleration data in the x-axis direction collected by the first accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the y-axis direction collected by the first accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the z-axis direction collected by the first accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the x-axis direction collected by the second accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the y-axis direction collected by the second accelerometer in the i-th group of signal data, is the numerical normalization value of the acceleration data in the z-axis direction collected by the second accelerometer in the i-th group of signal data; Similarly, numerically standardize the data of each subject; Step 1-3: Use the numerically standardized data corresponding to any one subject to determine the data whose correlation index with sitting posture classification is greater than the threshold, and use the determined data as the data related to sitting posture classification in each group; Step 1-4: Retain the data related to sitting posture classification in all groups, and then fuse the retained force-sensitive resistor data and accelerometer data in each group to obtain the fused data in each group; Take the fused data in each group as a row of the reference data matrix, and use the fused data in each group to form the reference data matrix; The specific process of the above Step 1-3 is as follows: For any sitting posture, the correlation index between the data collected by the r-th subject's first force-sensitive resistor and the acceleration data in the x-axis direction collected by the first accelerometer in this sitting posture is as follows: Among them, Then the correlation index matrix Q corresponding to the r-th subject in this sitting posture is: Calculate the absolute value of each element in matrix Q respectively, obtain all the elements in matrix Q whose absolute value is greater than 0.8, and then determine the data associated with this sitting posture according to all the obtained elements, that is, if the absolute value of r a,b is greater than 0.8, then both a and b are data associated with this sitting posture. Similarly, respectively determine the data related to each sitting posture of the r-th subject, and use all the determined data as the data related to sitting posture classification; Step 2: When the user sits on the seat cushion, use the force-sensitive resistors and accelerometers on the seat cushion to collect the user's signal data, and then select the data associated with the sitting posture classification from the collected user signal data; Step 3: Obtain the sitting posture classification result of the user according to the data selected in Step 2 and the reference data matrix.

2. The sitting posture classification method based on a seat cushion according to claim 1, wherein The sitting postures of the subject specifically include sitting upright, leaning left, leaning right, leaning forward, leaning backward, shaking the left leg, shaking the right leg, crossing the left leg, and crossing the right leg.

3. The sitting posture classification method based on a seat cushion according to claim 2, wherein, In Step 14, use the fused data groups to form a reference data matrix, specifically: Among them, n is the total number of data groups collected in Step 1, M is the reference data matrix, G is the number of data types associated with sitting posture classification, g = 1, 2, …, G, (x i1 , x i2 , …, x iG ) is the i-th group of data after fusion.

4. The sitting posture classification method based on a seat cushion according to claim 3, wherein, The specific process of Step 3 is: Step 3.1: Denote the data selected from the collected user signal data and associated with sitting posture classification as Step 32: Calculate respectively the correlation degree characteristic distance between each row vector in the reference data matrix M; where x i is the i-th row vector in the reference data matrix M, and x i =(x i1 , x i2 , …, x iG ), T represents the transpose operation, S is the covariance matrix, and S -1 is the inverse of the covariance matrix, is the correlation degree characteristic distance between Step 3: According to obtain the sitting posture classification result: Among them, C is the sitting posture classification set, is the sitting posture classification result of the user, I(y i = c) is the indicator function, y i is the true category of the i-th row vector in the reference data matrix M. When y i = c, the value of I(y i = c) is 1. When y i ≠ c, the value of I(y i = c) is 0.

5. The sitting posture classification method based on a seat cushion according to claim 4, characterized in that, The covariance matrix S is: where, σ jk is an element in the covariance matrix, j = 1, 2, …, G, k = 1, 2, …, G; Among them,

Images