JP2015097565A - Posture classification method, information processing apparatus, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To classify postures of human bodies with high accuracy.SOLUTION: Postures of human bodies are classified into a plurality of posture types. When a plurality of first measurement values 31, which measure the inclination of a subject 11 in a plurality of directions, are obtained for each posture type, a parameter calculation unit 3 calculates, for each posture type, parameters 21 for generation of a probability model 22 of measurement values that follows a multi-dimensional normal distribution on the basis of the plurality of first measurement values 31 for each posture type, and stores the parameters 21 in a storage unit 5. When second measurement values 32, which measure the inclination of the subject 11 in the plurality of directions each same as for the first measurement values 31, are also obtained, a classification processing unit 4 classifies the posture of the subject 11 into any of the plurality of posture types on the basis of the probability model 22 for each posture type generated based respectively on the second measurement values 32 and the parameters 21 stored in the storage device 5.

Description

  The present invention relates to a posture classification method, an information processing apparatus, and a program.

  It is known that the posture of the human body is related to low back pain, which accounts for 60% of occupational diseases. Therefore, it is considered to give a warning or guidance to a person who may develop low back pain by measuring the posture of the person at work.

  As a method for measuring the posture, for example, there is a method using a camera and a foot pressure sensor. However, this method cannot perform measurement outside a specific place and cannot monitor the posture on a daily basis. . Therefore, in recent years, a posture measuring method using a wearable sensor has been considered. For example, a plurality of parameters in which the posture of a person to be measured is determined in advance by obtaining a parameter using spectrum analysis based on a measurement value obtained by a triaxial acceleration sensor attached to a human body and comparing the parameter with a predetermined threshold value. There is a method of classifying into one of the classification items.

  Further, as an example of a technique related to posture estimation, there is one that recognizes a posture based on a captured image of a human body and corrects posture information corresponding to the posture based on a constraint condition based on the size of a face image. .

  Further, in an apparatus for generating a shape model of a model object whose shape is determined by a tip and a joint, such as a human hand, a constraint condition relating to a predetermined degree of freedom and rotation angle of the joint, position information of the tip, The angle of the joint linked to the tip is calculated based on the above.

JP 2010-125239 A JP 2011-253292 A JP 2002-92589 A

As described above, the method of comparing the parameter based on the measurement value by the acceleration sensor with a predetermined threshold has a problem that the posture cannot be classified with high accuracy.
In one aspect, an object of the present invention is to provide a posture classification method, an information processing apparatus, and a program capable of classifying a posture of a human body with high accuracy.

  In one proposal, a posture classification method is provided in which the following processing is performed. In this posture classification method, a plurality of first measurement values obtained by measuring the inclination of the human body with respect to a plurality of directions are acquired for each posture type in which the postures of the human body are classified into a plurality of types, and a plurality of first measurements for each posture type are obtained. Based on the value, a parameter for generating a probabilistic model of the measured value is calculated for each posture type, stored in the storage device, and the second measured by the human body tilt with respect to the same multiple directions as the first measured value. Based on the second measurement value and the probability model for each posture type generated based on the parameters stored in the storage device, the measurement target human body of the second measurement value is obtained. The processing includes classifying the posture into one of a plurality of types of posture.

  In one scheme, an information processing apparatus is provided. This information processing apparatus includes a measurement value acquisition unit and a parameter calculation unit. The measurement value acquisition unit acquires a measurement value obtained by measuring the inclination of the human body with respect to a plurality of directions. The parameter calculation unit obtains a plurality of first measurement values obtained by measuring the inclination of the human body with respect to a plurality of directions for each posture type obtained by classifying the postures of the human body into a plurality of types. Based on the first measurement value, a parameter for generating a probability model of the measurement value is calculated for each posture type and stored in the storage device.

  Furthermore, in one proposal, a program for causing a computer to execute the same processing as that of the information processing apparatus is provided.

  According to one aspect, the posture of the human body can be classified with high accuracy.

It is a figure which shows the structural example and processing example of the information processing apparatus which concern on 1st Embodiment. It is a figure which shows the structural example of the attitude | position measurement system which concerns on 2nd Embodiment. It is a figure which shows the hardware structural example of the attitude | position classification apparatus which concerns on 2nd Embodiment. It is a figure shown about the measured value by a sensor device. It is a figure which shows the example of the classification | category item of an attitude | position. It is a block diagram which shows the example of the function with which an attitude | position classification apparatus is provided. It is a flowchart which shows the example of the procedure of the whole process by an attitude | position classification apparatus. It is a flowchart which shows the example of the process sequence of the feature-value calculation in a learning process. It is a flowchart which shows the example of the process sequence of attitude | position parameter calculation. It is a flowchart which shows the example of the process sequence of the feature-value calculation in this process. It is a flowchart which shows the example of the process sequence of attitude | position classification. It is a figure for demonstrating the calculation for attitude | position classification. It is a reference figure in the case of performing posture classification by comparing a variable obtained by measurement with a rectangular determination region. It is a figure which shows the mounting position of a sensor device. It is a graph which shows the example of calculation of the correct answer rate according to the position of a sensor device. It is a figure which shows the structural example of the attitude | position measurement system which concerns on 3rd Embodiment. It is a block diagram which shows the structural example of the function with which the attitude | position classification apparatus which concerns on 3rd Embodiment is provided. It is a graph which shows the example of calculation of the correct answer rate in processing example 1. It is a figure which shows the example of the attitude | position at the time of inclining a back in the front-back direction. 12 is a flowchart illustrating an example of a processing procedure for posture parameter calculation in Processing Example 2; It is a graph which shows the example of calculation of the correct answer rate in processing example 2. It is a figure which shows the example of the attitude | position when a back inclines diagonally. It is a figure which shows the example of distribution of a posture angle in case the change of a posture is limited to the front-back direction. It is a figure which shows the example of distribution of an attitude | position angle in case an attitude | position changes also in the diagonal direction. 12 is a flowchart illustrating an example of a processing procedure of posture parameter calculation in Processing Example 3; 10 is a graph illustrating an example of calculating a correct answer rate in Processing Example 3; It is a graph which shows the example of calculation of the correct answer rate for every classification item. 14 is a flowchart illustrating an example of a procedure for calculating posture parameters in a processing example 4; It is a block diagram which shows the structural example of the terminal device which concerns on 4th Embodiment.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[First Embodiment]
FIG. 1 is a diagram illustrating a configuration example and a processing example of the information processing apparatus according to the first embodiment. An information processing apparatus 1 illustrated in FIG. 1 includes a measurement value acquisition unit 2, a parameter calculation unit 3, a classification processing unit 4, and a storage device 5. The processes of the measurement value acquisition unit 2, the parameter calculation unit 3, and the classification processing unit 4 are realized by, for example, a processor included in the information processing apparatus 1 executing a predetermined program. The storage device 5 is realized by a nonvolatile storage device such as an HDD (Hard Disk Drive). The storage device 5 may be provided outside the information processing device 1.

  The measurement value acquisition unit 2 acquires measurement values obtained by measuring the inclination of the human body (subject 11) with respect to a plurality of directions. The acquired measurement value is based on the measurement value output from the sensor 12 attached to the subject 11.

  Here, the measurement value acquired by the measurement value acquisition unit 2 may be the measurement value itself output from the sensor 12 or may be obtained by calculation based on the measurement value output from the sensor 12. It may be a thing. For example, when the sensor 12 is a three-axis acceleration sensor, the measurement value acquisition unit 2 calculates a measurement value indicating the inclination of the subject 11 with respect to a plurality of directions based on the acceleration measurement value output from the sensor 12.

For example, the sensor 12 is attached to the back of the subject 11.
The parameter calculation unit 3 calculates the parameter 21 used in the posture classification process in the classification processing unit 4 and stores it in the storage device 5. The classification processing unit 4 uses the parameter 21 stored in the storage device 5 to perform classification processing of the posture of the subject 11. Here, the parameter calculation unit 3 operates in a “learning step” before the posture classification processing in the classification processing unit 4.

  In the learning step, the measurement value acquisition unit 2 acquires a plurality of first measurement values 31 obtained by measuring the inclination of the human body with respect to a plurality of directions for each posture type classified into a plurality of types of human body postures. That is, a plurality of first measurement values 31 are acquired for each posture type.

  The posture types are for classifying posture states, and can be classified by states such as the back standing upright, the back leaning forward, and the back leaning backward. Further, in the learning step, when the first measurement value 31 corresponding to a certain posture type is acquired, the subject 11 takes a posture corresponding to the posture type, and measurement by the sensor 12 is performed in that state. .

  The parameter calculation unit 3 calculates the parameter 21 based on the first measurement value 31 acquired by the measurement value acquisition unit 2. Specifically, the parameter calculation unit 3 calculates a parameter 21 for generating a probability model 22 of measurement values based on a plurality of first measurement values 31 corresponding to a certain posture type. The probability model 22 is a probability model of measurement values according to a multidimensional normal distribution, for example. By executing such calculation processing for each posture type, a plurality of parameters 21 corresponding to the individual posture types are calculated and stored in the storage device 5.

  After the learning process is completed, a process for posture classification is performed. In the posture classification process, the measurement by the sensor 12 is performed again with the subject 11 in an arbitrary posture, and the measurement value acquisition unit 2 performs the second measurement in which the inclination of the subject 11 with respect to a plurality of directions is measured. Get the value 32. At this time, a plurality of second measurement values 32 may be acquired.

  Based on the acquired second measurement value 32 and the probability model 22 for each posture type generated based on the parameter 21 stored in the storage device 5, the classification processing unit 4 measures the subject 11 to be measured. Are classified into one of a plurality of posture types. That is, the classification processing unit 4 generates a probability model 22 corresponding to the posture type based on the parameter 21 corresponding to a certain posture type. In this way, the probability model 22 is generated for each posture type. The classification processing unit classifies the posture of the subject 11 to be measured into one of a plurality of posture types based on the probability model 22 generated for each posture type and the second measurement value 32.

  The probability model 22 is represented, for example, as a calculation formula for calculating the probability that the second measurement value 32 is classified into the corresponding posture type. In this case, the classification processing unit 4 calculates the probability for each posture type, and classifies the posture of the subject 11 into the posture classification having the highest calculated probability.

  According to the information processing apparatus 1 described above, parameters for generating a probability model 22 of measurement values based on a plurality of first measurement values 31 obtained by measuring the inclination of the subject 11 with respect to a plurality of directions for each posture type. 21 is calculated for each posture type. Then, when the second measurement value 32 is acquired, the posture of the subject 11 is any one based on the second measurement value 32 and the probability model 22 generated based on the calculated parameter 21. Classified into posture types. Thereby, the posture of the human body can be classified with high accuracy.

  In the above first embodiment, the subject 11 is the same in the learning step and the subsequent posture classification step, but the subject in each step may be a different person. Further, for example, the plurality of first measurement values 31 acquired for each posture type in the learning process may be obtained from a plurality of subjects.

[Second Embodiment]
FIG. 2 is a diagram illustrating a configuration example of an attitude measurement system according to the second embodiment. The posture measurement system 100 includes a sensor device 110, a terminal device 120, and a posture classification device 200. The terminal device 120 and the posture classification device 200 are connected via a network 130.

  The sensor device 110 is a so-called “wearable sensor” used by being attached to a human body. Specifically, the sensor device 110 is attached to the back of a human body. In addition, the sensor device 110 includes a triaxial acceleration sensor that detects triaxial acceleration based on the direction of gravity. Note that how the three-axis directions are defined will be described later.

  The terminal device 120 causes the sensor device 110 to perform acceleration measurement in accordance with, for example, an operation by the user. The terminal device 120 transmits the measurement value output from the sensor device 110 to the posture classification device 200 through the network 130.

  The posture classification apparatus 200 estimates the posture of the human body to be measured based on the measurement value transmitted from the terminal device 120 by the sensor device 110. Specifically, posture classification apparatus 200 classifies the posture of the human body to be measured into one of a plurality of predetermined posture classification items based on the measurement value by sensor device 110. In addition, posture classification apparatus 200 can also execute processing for calculating control parameters used during posture classification processing.

FIG. 3 is a diagram illustrating a hardware configuration example of the posture classification apparatus according to the second embodiment. The posture classification apparatus 200 is realized as a computer as shown in FIG. 3, for example.
The posture classification apparatus 200 is controlled by the processor 201 as a whole. The processor 201 may be a multiprocessor. The processor 201 is, for example, a central processing unit (CPU), a micro processing unit (MPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), or a programmable logic device (PLD). The processor 201 may be a combination of two or more elements among CPU, MPU, DSP, ASIC, and PLD.

A RAM (Random Access Memory) 202 and a plurality of peripheral devices are connected to the processor 201 via a bus 208.
The RAM 202 is used as a main storage device of the posture classification device 200. The RAM 202 temporarily stores at least part of an OS (Operating System) program and application programs to be executed by the processor 201. The RAM 202 stores various data necessary for processing by the processor 201.

  Peripheral devices connected to the bus 208 include an HDD 203, a graphic processing device 204, an input interface 205, a reading device 206, and a communication interface 207.

  The HDD 203 is used as an auxiliary storage device of the posture classification device 200. The HDD 203 stores an OS program, application programs, and various data. As the auxiliary storage device, other types of nonvolatile storage devices such as SSD (Solid State Drive) can be used.

  A display device 204 a is connected to the graphic processing device 204. The graphic processing device 204 displays an image on the screen of the display device 204a in accordance with an instruction from the processor 201. Examples of the display device include a display device using a CRT (Cathode Ray Tube) and a liquid crystal display device.

  An input device 205 a is connected to the input interface 205. The input interface 205 transmits a signal output from the input device 205a to the processor 201. Examples of the input device 205a include a keyboard and a pointing device. Examples of pointing devices include a mouse, a touch panel, a tablet, a touch pad, and a trackball.

  A portable recording medium 206a is attached to and detached from the reading device 206. The reading device 206 reads data recorded on the portable recording medium 206 a and transmits it to the processor 201. Examples of the portable recording medium 206a include an optical disk, a magneto-optical disk, and a semiconductor memory.

The communication interface 207 transmits / receives data to / from other devices such as the terminal device 120 via the network 130.
With the hardware configuration described above, the processing function of the posture classification apparatus 200 can be realized.

  Note that the terminal device 120 can also be realized as a computer having the same hardware configuration as in FIG. The terminal device 120 may be a portable information processing device such as a notebook personal computer or a tablet information terminal. In this case, the function of the sensor device 110 may be incorporated in the terminal device 120.

FIG. 4 is a diagram illustrating measurement values obtained by the sensor device.
As shown on the right side of FIG. 4, the sensor device 110 is attached to the back of the subject 140. Then, the sensor device 110 outputs a measurement value for estimating the posture of the upper body of the subject 140. As will be described later, the posture classification apparatus 200 classifies the back tilt and the curved state of the subject 140 based on the measurement values obtained by the sensor device 110. Note that the mounting position of the sensor device 110 is not limited to the back, and may be, for example, the shoulder or the neck, as long as the tilt and the curved state of the back of the subject 140 can be classified.

  As shown on the left side of FIG. 4, the sensor device 110 measures acceleration in each of the x direction, the y direction, and the z direction. The x direction, the y direction, and the z direction are detected as directions relative to the direction of gravity g.

  The x direction indicates a direction orthogonal to the front-rear direction with respect to the back of the subject 140, and the y direction indicates a direction orthogonal to the left-right direction with respect to the back of the subject 140. In the present embodiment, as shown in FIG. 4, the rear direction of the subject 140 is the positive direction for the x direction, and the right direction of the subject 140 is the positive direction for the y direction.

  The z direction indicates a direction along the back of the subject 140. In the present embodiment, since the measurement is performed with the upper body of the subject 140 standing, the z direction is perpendicular to or close to the ground. Further, in the present embodiment, as shown in FIG. 4, the upward direction is the positive direction with respect to the z direction.

  FIG. 5 is a diagram illustrating an example of posture classification items. The posture classification apparatus 200 determines which of the 13 types of classification items shown in FIG. These classification items are classified into five major classifications and three minor classifications.

  The major classification classifies the inclination of the back of the subject, and is classified into five types: “upright”, “forward tilt”, “back tilt”, “right tilt”, and “left tilt”. “Upright” indicates that the subject's back is in an upright state and is not inclined in the front-rear direction or the left-right direction. “Forward tilt” indicates that the subject's back is tilted forward. “Backward tilt” indicates that the subject's back is tilted rearward. “Tilt right” indicates that the subject's back is tilted to the right. “Left tilt” indicates that the subject's back is tilted to the left.

  The small classification is for classifying the curved state of the subject's back, and is classified into three types: “straight”, “warping”, and “back of the back”. “Straight” indicates that the subject's back is straight and not curved. “Warpage” indicates that the subject's back is curved (ie, the back side of the spine is recessed). “Stoopback” indicates that the subject's back is curved (ie, the back side of the spine protrudes).

  As an overall classification item, each of “upright”, “forward tilt”, and “backward tilt” is further classified into three types of “straight”, “warp”, and “stoop”. Each of “right tilt” and “left tilt” is further classified into only two types, “straight” and “stoop”. As a result, the posture of the subject is classified into a total of 13 types of classification items.

FIG. 6 is a block diagram illustrating an example of functions provided in the posture classification apparatus.
The posture classification apparatus 200 includes a data acquisition unit 211, a feature amount calculation unit 212, a posture parameter calculation unit 213, a posture parameter storage unit 214, a posture classification processing unit 215, and a correct answer rate calculation unit 216. For each processing of the data acquisition unit 211, the feature amount calculation unit 212, the posture parameter calculation unit 213, the posture classification processing unit 215, and the correct answer rate calculation unit 216, for example, the processor 201 included in the posture classification device 200 executes a predetermined program. This is realized. The posture parameter storage unit 214 is realized as a storage area of the HDD 203 provided in the posture classification apparatus 200, for example.

  The data acquisition unit 211 acquires a measurement value obtained by the sensor device 110 from the terminal device 120. The acquired measurement value is the acceleration in each of the x direction, the y direction, and the z direction.

  The feature amount calculation unit 212 calculates a feature amount related to the posture based on the acquired measurement value. As the feature amount, a “posture angle” indicating each inclination angle in the left-right direction and the front-rear direction of the back is calculated.

  The posture parameter calculation unit 213 calculates a “posture parameter” used in the posture classification process based on the calculated feature amount, and stores it in the posture parameter storage unit 214. As the posture parameter, an average vector and a covariance matrix based on the posture angle are calculated. In addition, the posture parameter is calculated for each of the 13 types of classification items described above.

  The posture classification processing unit 215 refers to the posture parameters stored in the posture parameter storage unit 214, and based on the feature amount calculated by the feature amount calculation unit 212, the posture of the subject is classified into any of the above classification items. Judge whether or not. Specifically, the posture classification processing unit 215 calculates the probability that the posture of the subject belongs to each classification item by calculating the calculated feature value and the posture parameter corresponding to each of the 13 types of classification items. Ask for. Then, the posture classification processing unit 215 outputs the classification item having the highest obtained probability as the classification item indicating the posture of the subject.

  Note that the posture parameter calculation unit 213 is used only in preprocessing for posture parameter calculation, which is executed before processing for actually classifying the posture of the subject. On the other hand, the posture classification processing unit 215 is used when actually classifying the posture of the subject. Therefore, the feature amount acquired by the posture parameter calculation unit 213 from the feature amount calculation unit 212 and the feature amount acquired by the posture classification processing unit 215 from the feature amount calculation unit 212 are based on measured values measured at different timings. It is.

  The correct answer rate calculation unit 216 calculates the correct answer rate of the classification result obtained by the posture classification processing unit 215. The correct rate calculation unit 216 is a processing function provided to verify the accuracy of the posture classification process. Therefore, for example, when the posture classification device 200 is shipped as a product, the posture classification device 200 may not include the correct answer rate calculation unit 216.

  FIG. 7 is a flowchart illustrating an example of a procedure of the entire process performed by the posture classification apparatus. Processing by the posture classification apparatus 200 is roughly divided into “learning processing” and “main processing”. This process is a process of actually classifying the posture of the subject based on the posture parameter. On the other hand, the learning process is a process executed to calculate the posture parameters used in the present process prior to the present process. As described above, the posture parameter is calculated for each of the 13 types of classification items.

The learning process is executed in the following procedure. In the following description, “j” is an integer of 1 or more, and indicates the number of classification items. In this embodiment, as an example, j = 13.
[Step S11] The subject is assumed to belong to a certain classification item, and measurement by the sensor device 110 is performed in that state. Further, such measurement is performed N times (where N is an integer of 1 or more).

  Note that each of the N measurements is basically performed on the same subject. However, each measurement in N measurements may be performed using two or more subjects.

  The data acquisition unit 211 acquires the measurement results for N times by the sensor device 110. The feature amount calculation unit 212 calculates a feature amount, that is, a posture angle indicating each inclination angle in the left-right direction and the front-rear direction of the back, based on the acquired measurement result.

  [Step S12] The posture parameter calculation unit 213 calculates a posture parameter corresponding to one posture classification item based on the calculated feature amount, and stores the posture parameter in the posture parameter storage unit 214. As described above, the posture parameter includes an average vector and a covariance matrix based on the posture angle.

  [Step S13] It is determined whether or not the processing of steps S11 and S12 has been executed for all of the j types of posture classification items. If there is a posture classification item that has not been processed, the process returns to step S11, and the processing of steps S11 and S12 is executed for another posture classification item that has not been processed. On the other hand, if all the posture classification items have been processed, the learning process is terminated.

Through the above processing, posture parameters are calculated for each of j (= 13) types of classification items, and the calculated posture parameters are stored in the posture parameter storage unit 214.
Next, in this process, the following procedure is executed.

  [Step S21] A measurement result is output from the sensor device 110 attached to the subject. Basically, the subject in this process is the same as the subject in the learning process, but the subject in the learning process and the subject in this process can be different.

  The data acquisition unit 211 acquires the measurement results for N times by the sensor device 110. The feature amount calculation unit 212 calculates a feature amount, that is, a posture angle indicating each inclination angle in the left-right direction and the front-rear direction of the back, based on the acquired measurement result.

  [Step S22] The posture classification processing unit 215 refers to the posture parameters stored in the posture parameter storage unit 214, and based on the feature amount calculated by the feature amount calculation unit 212, the subject's posture is j types of classification items. It is determined to be classified into.

Next, details of the processing of FIG. 7 will be described.
First, FIG. 8 is a flowchart illustrating an example of a processing procedure for calculating a feature amount in the learning process. The processing in FIG. 8 corresponds to the processing in step S11 in FIG.

  [Step S101] The subject is allowed to take a posture that belongs to a certain classification item, and measurement by the sensor device 110 is performed in that state. Here, as an example, the subject is kept stationary for t seconds (where t> 0) in a posture that belongs to a certain classification item. Then, the sensor device 110 is caused to perform measurement at intervals of 1 second in a stationary period of t seconds, and t measurement results are output. For example, t is “15” (that is, the stationary period is 15 seconds).

The measurement results are expressed as the following formulas (1-1) to (1-3). In Expressions (1-1) to (1-3), X _jN , Y _jN , and Z _jN indicate the measured values in the Nth x-direction, y-direction, and z-direction corresponding to the j-th classification item, respectively.
X _jN = {X _jN (1), X _jN (2),..., X _jN (t)} (1-1)
Y _jN = {Y _jN (1), Y _jN (2),..., Y _jN (t)} (1-2)
Z _jN = {Z _jN (1), Z _jN (2),..., Z _jN (t)} (1-3)
The data acquisition unit 211 acquires the measurement values X _jN , Y _jN , and Z _jN via the terminal device 120.

[Step S102] The feature amount calculation unit 212 calculates an average value of the measurement values X _jN , Y _jN , and Z _jN in the measurement period of t seconds. The average values Ave (X _jN ), Ave (Y _jN ), and Ave (Z _jN ) of the measured values X _jN , Y _jN , and Z _jN are calculated according to the following equations (2-1) to (2-3). Is done.

[Step S103] The feature amount calculation unit 212 calculates the respective inclination angles of the back in the left-right direction and the front-rear direction based on the average values Ave (X _jN ), Ave (Y _jN ), and Ave (Z _jN ). The tilt angle calculated here is expressed as an angle based on the direction of gravity. Lateral direction of the inclination angle phi _JN1, longitudinal inclination angle phi _JN2, the following formula, respectively (3-1) is calculated according to (3-2).

[Step S104] The feature amount calculation unit 212 calculates the posture angle by correcting the tilt angles φ _jN1 and φ _jN2 calculated in step S103. This correction is for reducing an error due to the sensor device 110 being attached to the subject in a tilted state. Here, assuming that the y direction of the sensor device 110 is tilted, that is, when the sensor device 110 is mounted in a tilted state when viewed from the rear side of the subject, the tilt angle in the left and right direction is assumed. _Correct φ _jN1 .

When the tilt angle in the left-right direction of the sensor device 110 is expressed as φ _{jN01 in the} reference posture (that is, when the sensor device 110 is mounted in an ideal state), the posture angles θ _jN1 , θ _{jN2 in the} left-right direction and the front-back direction are _given. Is calculated according to the following equation (4).
(Θ _jN1 , θ _jN2 ) = (φ _jN1 −φ _jN01 , φ _jN2 ) (4)
[Step S105] It is determined whether the number of measurements, that is, the number of executions of steps S101 to S104, has reached N times. If the number of measurements is less than N, the process returns to step S101 and the measurement is continued. If the number of measurements reaches N, the process proceeds to step S12 in FIG. Through the processes in steps S101 to _S105 , the posture angles θ _jN1 and θ _jN2 corresponding to one classification item are calculated.

Next, FIG. 9 is a flowchart illustrating an example of a processing procedure for calculating posture parameters. The process of FIG. 9 corresponds to the process of step S12 of FIG. In the following description, the posture angle (θ _jN1 , θ _jN2 ) is represented as (θ _j1 (N), θ _j2 (N)).

[Step S111] The posture parameter calculation unit 213 calculates an average vector based on the posture angle.
Here, the posture angle Θ _j calculated for the j-th classification item is expressed as the following equation (5). Further, average values Ave (θ _j1 ) and Ave (θ _j2 ) of the posture angles in the left-right direction and the front-rear direction in the N measurements are expressed by equations (6-1) and (6-2), respectively. The subscript T indicates a transposed matrix.

Then, the average vector Ave (Θ _j ) corresponding to the jth classification item is calculated according to the following equation (7).
Ave (Θ _j ) = (Ave (θ _j1 ), Ave (θ _j2 )) ^T (7)
[Step S112] The attitude parameter calculation unit 213 calculates a covariance matrix based on the attitude angle and the average vector. Here, assuming that Bj = Θ _j −Ave (Θ _j ), the covariance matrix A _j corresponding to the j-th classification item is calculated according to the following equation (8).
A _j = B _j B _j ^T (8)
The average vector Ave (Θ _j ) is a two-dimensional vector, and the covariance matrix A _j is a 2 × 2 matrix. Through the above processing, the posture parameter corresponding to the jth classification item is calculated. The posture parameter calculation unit 213 stores the calculated posture parameters, that is, the average vector Ave (Θ _j ) and the covariance matrix A _j in the posture parameter storage unit 214.

Next, FIG. 10 is a flowchart illustrating an example of a processing procedure for calculating a feature amount in this processing. The processing in FIG. 10 corresponds to the processing in step S21 in FIG.
The processing procedures of steps S201, S202, S203, and S204 shown in FIG. 10 are the same as steps S101, S102, S103, and S104 of FIG. That is, the feature amount calculation process in this process is a modification of the process in FIG. 8 so that the measurement process for t seconds by the sensor device 110 is executed only once with the number of times of measurement N = 1. However, the subject in this process is different from the subject in the learning process. Note that a plurality of measurement processes for t seconds may be executed in the feature amount calculation process in this process.

The posture angles θ _jN1 and θ _{jN2 in} the left-right direction and the front-rear direction are calculated by the processing of FIG. 10 by the data acquisition unit 211 and the feature amount calculation unit 212. Hereinafter, the posture angles θ _jN1 and θ _jN2 calculated in this process will be described as θ ₁ and θ ₂ , respectively.

FIG. 11 is a flowchart illustrating an example of a processing procedure for posture classification. The process of FIG. 11 corresponds to the process of step S22 of FIG.
[Step S211] The posture classification processing unit 215 selects one of the j types of classification items, and acquires the posture parameters (average vector and covariance matrix) corresponding to the selected classification item from the posture parameter storage unit 214. get. Hereinafter, the average vector and the covariance matrix corresponding to the acquired j-th classification item are Ave (Θ _j ) and A _j , respectively.

[Step S212] The posture classification processing unit 215 receives the average vector Ave (Θ _j ) and covariance matrix A _j acquired from the posture parameter storage unit 214, and the posture angle θ = (θ ₁₎ calculated by the feature amount calculation unit 212. , Θ ₂ ), the probability P _j (θ) is calculated according to the following equation (9).

In equation (9), d represents the number of dimensions. This d coincides with the number of parameters of the posture angle θ, and d = 2 in the present embodiment.
Equation (9) above indicates the probability that the posture angle θ indicating the posture of the subject belongs to the classification item selected in step S211.

  [Step S213] The posture classification processing unit 215 determines whether all of the j types of classification items have been selected in Step S211. If there is an unselected classification item, the process of step S211 is executed. In this case, the processes of steps S211 and S212 are executed for the unselected classification item. That is, by repeating steps S211 and S212, the probability that the posture angle θ representing the posture of the subject belongs to each of n types of classification items is calculated. On the other hand, if all the classification items have been selected, the process of step S214 is executed.

  [Step S214] The posture classification processing unit 215 determines that the classification item corresponding to the probability having the highest value among the calculated n probabilities is the classification item to which the posture of the subject belongs and outputs the classification item.

Here, FIG. 12 is a diagram for explaining calculation for posture classification. The graph of FIG. 12 shows a two-dimensional normal distribution on a plane by two variables, that is, a posture angle in the left-right direction (corresponding to θ _jN1 ) and a posture angle in the front-rear direction (corresponding to θ _jN2 ).

  In addition, elliptical areas 301, 302, and 303 in FIG. 12 correspond to different classification items # 1, # 2, and # 3, respectively. The ellipse area 301 is an area in which two variables obtained from a subject whose posture belongs to the classification item # 1 are included at a certain ratio. Similarly, the elliptical region 302 is a region in which two variables obtained from the subject who is in the posture belonging to the classification item # 2 are included at a certain ratio. Similarly, the elliptical region 303 is a region in which two variables obtained from the subject who is in the posture belonging to the classification item # 3 are included at a certain ratio. Such ellipse regions 301-303 are called probability ellipsoids, a certain percentage of confidence ellipses, or probability deviation ellipses.

  In addition, points 311 and 312 in FIG. 12 indicate positions where two variables obtained from a subject having an arbitrary posture in this processing are plotted on the graph. Here, the above-described equation (9) represents a probability density function of a multidimensional normal distribution (a two-dimensional normal distribution in the present embodiment). This equation (9) calculates the distance between the points plotted on the graph of FIG. 12 and the center points of the elliptical regions 301 to 303 from the two variables obtained from the subject in any posture in this process. Is equivalent to Then, the posture classification processing unit 215 determines an elliptical region having the closest distance between the point plotted on the graph and the center point, and determines that the classification item corresponding to the determined elliptical region belongs to the subject's posture. To do.

  For example, the point 311 is closest to the central point of the elliptical region 301 among the central points of the elliptical regions 301 to 303. For this reason, it is determined that the posture of the subject from which the point 311 is obtained belongs to the classification item # 1. Further, the point 312 is closest to the center point of the elliptical region 302 among the central points of the elliptical regions 301 to 303. For this reason, it is determined that the posture of the subject from which the point 312 is obtained belongs to the classification item # 2.

  FIG. 13 is a reference diagram when posture classification is performed by comparing a variable obtained by measurement with a rectangular determination region. The posture classification method shown in FIG. 13 is compared with the posture classification method shown in FIG.

  In the graph of FIG. 13, rectangular areas 321 to 323 indicating upper and lower thresholds of two variables are set on a plane with two variables as axes. The rectangular areas 321, 322, and 323 are areas for determining that they belong to the classification items # 1, # 2, and # 3, respectively.

  Further, points 331 and 332 in FIG. 13 indicate positions where two variables obtained from a subject in an arbitrary posture in this processing are plotted on the graph. In the method of FIG. 13, the posture of the subject from which each point is obtained is classified according to which rectangular area each point is included in. However, in the example of FIG. 13, the point 331 is not included in any of the rectangular areas 321 to 323, and the point 332 is not included in any of the rectangular areas 321 to 323. In this case, it cannot be determined to which classification item the posture of the subject from which the points 331 and 332 are obtained belongs.

  On the other hand, according to the method shown in FIG. 12, the ellipse area where the distance between the point plotted on the graph and the center point is the closest is determined, and the classification item corresponding to the determined ellipse area is the subject. It is determined that the posture belongs. This makes it possible to reliably determine to which classification item the posture of the subject from whom the plotted points are obtained belongs. Even when the number of classification items is increased, it is possible to determine which classification item the subject's posture belongs to.

Next, verification of posture classification accuracy using the accuracy rate calculation unit 216 will be described. The inventors of the present application verified posture classification accuracy by the following method using 10-fold cross-validation.
First, the subject is assumed to belong to one classification item, and measurement is performed for t (= 15) seconds by the sensor device 110 in that state. Based on the measurement values obtained in this way, the feature amount, that is, the posture angle is obtained. Moreover, such a measurement is performed 10 times to obtain the posture angle. That is, ten posture angles are obtained for one classification item. Further, the above processing is repeated for each classification item, and 10 posture angles are obtained for each classification item.

  Next, 10 sets of data groups each including one posture angle for all classification items are created, and the posture angles are classified. Then, learning processing is performed using nine data groups out of the ten data groups, posture parameters for each classification item are calculated, and posture angles included in the remaining one data group are used. This process, that is, which classification item is determined is determined. In this case, learning processing is performed with N = 9, and it is determined to which classification item each of the ten posture angles included in the remaining data group belongs.

  Further, the learning process and the main process are repeated while changing the data group used for the main process. As a result, the posture classification is determined 100 times in total. The accuracy rate calculation unit 216 determines whether the determination result is correct each time the determination result of posture classification is obtained, and calculates the accuracy rate.

  However, in the calculation example of the accuracy rate shown below, only for the classification item “upright-straight”, measurement was performed 60 times instead of 10 times. Each data group includes six posture angles corresponding to the classification item “upright-straight”.

Further, as shown in FIG. 14, the processing was performed while changing the mounting position of the sensor device 110. FIG. 14 is a diagram illustrating a mounting position of the sensor device.
As shown in FIG. 14, the inventors attached the sensor device 110 to a position shifted in the vertical direction along the back of the subject 140 and verified posture classification. Hereinafter, the sensor devices 110 attached to the respective positions will be referred to as “sensor # 1”, “sensor # 2”, and “sensor # 3” in order from the top. Sensor # 1 is assumed to be worn near the scapula of subject 140. Sensor # 2 is assumed to be worn near the center of the back of subject 140. It is assumed that sensor # 3 is attached near the lower belt of the subject 140.

  FIG. 15 is a graph illustrating an example of calculating the correct answer rate according to the position of the sensor device. When the sensor # 1 is used, the accuracy rate is about 0.73, when the sensor # 2 is used, the accuracy rate is about 0.65, and when the sensor # 3 is used, the accuracy rate is 0.89. A correct answer rate of 60% or more was obtained at any position where 110 was attached. In particular, when sensor # 3 was used, a high accuracy rate of 80% or more was obtained. The correct answer rate for each classification item will be described later.

[Third Embodiment]
By the way, in said 2nd Embodiment, as shown in FIG. 15, the variation in a correct answer rate is seen by the mounting position of the sensor device 110. As shown in FIG. On the other hand, it is considered that the accuracy of posture classification can be improved by performing learning processing and main processing using measurement values from a plurality of sensor devices 110 mounted at different positions.

  Therefore, in the following third embodiment, the posture classification apparatus 200 according to the second embodiment is modified so as to perform processing based on the measurement values of the plurality of sensor devices 110. Furthermore, the third embodiment also discloses means for correcting posture parameters calculated using measured values of a plurality of sensor devices 110 in order to obtain a further improvement effect of posture classification accuracy.

In the following description of the third embodiment, the same reference numerals are given to components and processing steps in which the same processing as in the second embodiment is performed, and description thereof is omitted.
FIG. 16 is a diagram illustrating a configuration example of an attitude measurement system according to the third embodiment. The posture measurement system 100a illustrated in FIG. 16 includes a plurality of sensor devices 110, a terminal device 120a, and a posture classification device 200a. The terminal device 120a and the attitude classification device 200a are connected via the network 130.

  The terminal device 120a is different from the terminal device 120 illustrated in FIG. 2 in that the terminal device 120a is connected to a plurality of sensor devices 110 and transmits measurement values from the respective sensor devices 110 to the posture classification device 200a. In the example of FIG. 16, two sensor devices 110 are connected to the terminal device 120a, but three or more sensor devices 110 may be connected.

  The plurality of sensor devices 110 are worn by the same subject. In addition, it is desirable that the plurality of sensor devices 110 be mounted at different positions in the vertical direction, that is, the direction along the spine, on the subject's back or in the vicinity thereof. For example, when two sensor devices 110 are used, these sensor devices 110 may be any of sensor # 1 and sensor # 2, sensor # 1 and sensor # 3, sensor # 2 and sensor # 3 shown in FIG. You may make it respond | correspond to these combinations.

Note that the functions of the plurality of sensor devices 110 may be incorporated in the terminal device 120a, for example.
The posture classification apparatus 200a performs posture parameter calculation and posture classification processing using the measurement values received from each of the plurality of sensor devices 110.

  FIG. 17 is a block diagram illustrating a configuration example of functions included in the posture classification apparatus according to the third embodiment. The posture classification apparatus 200a illustrated in FIG. 17 includes a data acquisition unit 211a, a feature amount calculation unit 212a, a posture parameter calculation unit 213a, a posture parameter storage unit 214a, a posture classification processing unit 215a, a correct answer rate calculation unit 216, and a constraint condition calculation unit 221. And a posture parameter correction unit 222. Note that the hardware configuration of the posture classification apparatus 200a can be the same as that of the posture classification apparatus 200 shown in FIG.

  Each process of the data acquisition unit 211a, the feature amount calculation unit 212a, the posture parameter calculation unit 213a, the posture classification processing unit 215a, the accuracy rate calculation unit 216, the constraint condition calculation unit 221 and the posture parameter correction unit 222 is performed by, for example, a posture classification device. This is realized by the processor 201 included in the 200a executing a predetermined program. In addition, the posture parameter storage unit 214a is realized as a storage area of the HDD 203 provided in the posture classification apparatus 200a, for example.

  The basic processes of the data acquisition unit 211a, the feature amount calculation unit 212a, the posture parameter calculation unit 213a, and the posture classification processing unit 215a are the data acquisition unit 211, the feature amount calculation unit 212, and the posture parameter calculation unit illustrated in FIG. 213 and the posture classification processing unit 215 are the same. In the following, these processing functions will be described only with respect to differences from the corresponding processing functions in FIG.

  The data acquisition unit 211a is different from the data acquisition unit 211 in that it acquires acceleration measurement values for each of the x direction, the y direction, and the z direction from each of the two sensor devices 110.

  The feature amount calculation unit 212a is different from the feature amount calculation unit 212 in that a posture angle is calculated for each sensor device 110 using measurement values acquired from the two sensor devices 110, respectively.

  The posture parameter calculation unit 213a is different from the posture parameter calculation unit 213 in that a set of average vector and covariance matrix is calculated based on the posture angle calculated for each sensor device 110. In addition, when the constraint condition calculation unit 221 determines that there is a parameter with a small contribution to the posture classification, the posture parameter calculation unit 213a has a parameter with a small contribution, that is, one of the postures in the left-right direction and the front-rear direction. The attitude parameter is calculated excluding the angle.

  The posture classification processing unit 215a is different from the posture classification processing unit 215 in that the dimension d in the posture classification calculation formula is 2n, where n is the number of sensor devices 110. In addition, depending on the processing result of the constraint condition calculation unit 221, the dimension d may be smaller than 2n.

  The constraint condition calculation unit 221 and the posture parameter correction unit 222 are processing functions added to optimize posture parameters calculated based on the measurement values of the two sensor devices 110.

  The constraint condition calculation unit 221 determines a parameter having a small contribution to the posture classification process in consideration of the constraint condition related to the movement of the spine of the human body. There are the following two types of determination methods.

  As a first determination method, the constraint condition calculation unit 221 determines the direction of the back inclination with a small contribution to the posture classification process based on the feature amount calculated by the feature amount calculation unit 212a. In this case, the posture parameter calculation unit 213a calculates the posture parameter without using the feature amount corresponding to the determined direction, that is, the posture angle.

  Further, as a second determination method, the constraint condition calculation unit 221 determines a parameter that has a small contribution to the posture classification process based on the posture parameter calculated by the posture parameter calculation unit 213a. In this case, the constraint condition calculation unit 221 converts the posture parameter calculated by the posture parameter calculation unit 213a so as to exclude a parameter having a small contribution to the posture classification process, and passes the posture parameter to the posture parameter correction unit 222.

  The posture parameter correction unit 222 recalculates the posture parameter using the parameter converted by the constraint condition calculation unit 221 when the constraint condition calculation unit 221 performs the determination by the second determination method. The posture parameter correction unit 222 stores the recalculated posture parameter in the posture parameter storage unit 214a.

Hereinafter, the processing of the posture classification apparatus 200a will be described as processing examples of four patterns of processing examples 1 to 4.
<Processing Example 1>
The processing example 1 is a modification in which the processing of the posture classification apparatus 200 according to the second embodiment is performed using the measurement values of the plurality of sensor devices 110. In the processing example 1, the constraint condition calculation unit 221 and the posture parameter correction unit 222 are not used.

  The basic processing flow in Processing Example 1 is the same as that in FIG. 7 and includes a learning process and a main process. First, the feature amount calculation process (corresponding to step S11 in FIG. 7 and FIG. 8) in the learning process is modified as follows.

The data acquisition unit 211 a acquires measurement values obtained by the plurality of sensor devices 110 through the terminal device 120. _Assuming that the number of sensor devices 110 is n, the data acquisition unit 211a acquires n sets of measurement values X _jN , Y _jN , and Z _jN represented by the above formulas (1-1) to (1-3). . This process corresponds to step S101 in FIG. The feature quantity calculation unit 212a calculates n sets of posture angles θ _jN1 and θ _jN2 based on each of the acquired n sets of measurement values in the same procedure as steps _S102 to _S104 in FIG.

Further, the above n sets of posture angles θ _jN1 and θ _jN2 are calculated N times. As a result, n groups of data including N posture angles θ _jN1 and θ _jN2 are calculated as feature amounts corresponding to one classification item.

Next, the posture parameter calculation process (corresponding to step S12 in FIG. 7 and FIG. 9) is modified as follows.
The posture parameter calculation unit 213a calculates a set of average vector and covariance matrix based on the posture angle calculated for each sensor device 110. This process corresponds to step S111 in FIG.

Here, assuming that the posture angle calculated based on the n-th posture angle, that is, the measured value of the nth sensor device 110, is θ _{j1 — n} (N), θ _{j2 — n} (N), the j th The posture angle Θ _j calculated for the classification item is expressed as the following equation (5 ′). In addition, average values Ave (θ _{j1 — n} ) and Ave (θ _{j2 — n} ) of posture angles in the left-right direction and the front-rear direction in N measurements using the n-th sensor device 110 are expressed by the following formulas (6 '-1) and (6'-2). The average values Ave (θ _j1 ) and Ave (θ _j2 ) of the posture angles in the left-right direction and the front-rear direction in consideration of the measured values of all the sensor devices 110 are expressed by the following equations (6′-3) and (6 Represented by '-4).

Therefore, the average vector Ave (Θ _j ) corresponding to the j-th classification item is obtained by adding the average value of the posture angles represented by the equations (6′-3) and (6′-4) to the equation (7) described above. It is calculated by substituting Ave (θ _j1 ) and Ave (θ _j2 ).

Further, the covariance matrix A _j includes the attitude angle Θ _j calculated by the equation (5 ′) and the average vector Ave (Θ _j ) calculated by the above procedure using the attitude angle Θ _j . It is calculated by substituting into equation (8). This process corresponds to step S112 in FIG.

In the second embodiment, the average vector Ave (Θ _j ) is a two-dimensional vector, and the covariance matrix A _j is a matrix with 2 rows and 2 columns. On the other hand, in the processing example 1 of the third embodiment, the average vector Ave (Θ _j ) is a 2n-dimensional vector, and the covariance matrix A _j is a matrix of 2n rows and 2n columns.

As described above, the process obtained by modifying steps S11 and S12 in FIG. 7 is executed for each of the j types of classification items.
Next, the feature amount calculation process (corresponding to step S21 in FIG. 7 and FIG. 10) in this process is modified as follows.

The data acquisition unit 211 acquires measurement values obtained by the plurality of sensor devices 110 through the terminal device 120. That is, the data acquisition unit 211 acquires n sets of measurement values X _jN , Y _jN , and Z _jN represented by the above formulas (1-1) to (1-3). This process corresponds to step S201 in FIG. The feature amount calculation unit 212a calculates n sets of posture angles θ _jN1 and θ _jN2 based on each of the acquired n sets of measurement values in the same procedure as steps _S202 to _S204 in FIG.

Next, the posture classification process (corresponding to step S22 in FIG. 7 and FIG. 11) is modified as follows.
The posture classification processing by the posture classification processing unit 215a is equivalent to the processing in FIG. 11 performed by the posture classification processing unit 215 in which the processing in step S212 is modified as follows. As described above, the feature amount calculation unit 212a calculates n sets of posture angles θ _jN1 and θ _jN2 . Here, n th sensor device 110 measurements attitude angle is calculated based on _θ jN1, θ jN2 respectively θ ₁ _ _n, when the θ ₂ _ _n, is calculated attitude angle theta, the following formula It is expressed as (10).

The posture classification processing unit 215a calculates the probability P _j (θ) by substituting the posture angle θ represented by the equation (10) into the above-described equation (9). Note that the dimension d in equation (9) is 2n.

The posture classification processing unit 215a executes the calculation of the probability P _j (θ) as described above for each of the n types of classification items. Then, the posture classification processing unit 215a determines and outputs the classification item corresponding to the probability having the highest value among the calculated n probabilities as the classification item to which the posture of the subject belongs.

  Here, FIG. 18 is a graph showing an example of calculating the correct answer rate in Processing Example 1. FIG. 18 shows an example of the correct answer rate calculated by the correct answer rate calculation unit 216 when processing is performed with n = 2. FIG. 18 also shows combinations of mounting positions of the two sensor devices 110, that is, combinations of sensor # 1 and sensor # 2, sensor # 1 and sensor # 3, and sensor # 2 and sensor # 3 shown in FIG. The correct answer rate is shown when processing is performed. Further, in FIG. 18, for reference, the accuracy rate when processed by the posture classification apparatus 200 of the second embodiment using the sensors # 1, # 2, and # 3 alone (shown in FIG. 15). (Accuracy rate).

  As shown in FIG. 18, it has been found that when processing is performed by combining a plurality of sensor devices 110, the accuracy of posture classification may be lower than when processing is performed using the sensor device 110 alone. The inventors inferred the cause of this as follows.

  From a physical point of view, when there are two sensor devices 110 to be mounted, for example, the orientation parameter dimension expressing the orientation is “4”. On the other hand, there are some constraints on the movement of the bones that make up the spine of the human body. For this reason, the dimension of the posture parameter contributing to the posture classification may be less than “4”. In this case, it is conceivable that the classification accuracy deteriorates when the classification process is performed including a dimension parameter that does not contribute to the posture classification.

  From a numerical point of view, it is also conceivable that the calculation accuracy using the inverse matrix has deteriorated due to the presence of extremely small eigenvalues in the eigenvalues of the covariance matrix compared to the maximum eigenvalue. Such deterioration in accuracy may also be attributed to the constraint of bone movement.

The following processing examples 2 and 3 are modifications of the above processing example 1 so that posture parameters used in posture classification are optimized based on such a cause inference.
<Processing example 2>
FIG. 19 is a diagram illustrating an example of a posture when the back is tilted in the front-rear direction. The left side of FIG. 19 shows an example of the posture when the subject 140 tilts his back forward, and the right side of FIG. 19 shows an example of the posture when the subject 140 tilts his back backward.

  As in the example of FIG. 19, when the subject 140 performs an operation of tilting the back in the front-rear direction, the back may hardly tilt in the left-right direction. In this case, the value of the parameter indicating the tilt angle in the left-right direction hardly changes depending on whether the subject 140 tilts the back in the forward direction, tilts in the rear direction, or in the middle. Therefore, for example, in order to classify postures with different back-and-forth back-and-forth orientations, such as “upright”, “forward tilt”, and “backward tilt” among the classification items described above, the tilt angle in the left-right direction is The parameters shown are unlikely to contribute.

  As described above, if a posture parameter includes a parameter related to a direction that does not contribute to posture classification, the parameter may adversely affect posture classification processing. Therefore, in the processing example 2, the constraint condition calculation unit 221 determines whether there is a direction that does not contribute to the posture classification process among the directions of the inclination angle indicated by the posture parameter (that is, the left-right direction or the front-back direction). If there is a direction that does not contribute to the posture classification process, the parameter relating to the direction is discarded and is not used for the posture classification process.

  FIG. 20 is a flowchart illustrating an example of a processing procedure of posture parameter calculation in Processing Example 2. In the processing example 2, the posture parameter calculation processing (corresponding to step S12 in FIG. 7 and FIG. 9) is modified as follows.

  [Step S151] The constraint condition calculation unit 221 determines whether there is a parameter corresponding to a direction that does not contribute to the posture classification process based on the posture angle calculated by the feature amount calculation unit 212a.

Specifically, the constraint condition calculation unit 221 calculates the total posture angle indicating the tilt angle in the same direction. In this calculation, the posture angles corresponding to the same direction based on the measurement values from the sensor devices 110 are summed. _Assuming that the posture angles calculated based on the N measurement values by the n-th sensor device 110 are θ _{j1 — n} (N) and θ _{j2 — n} (N), the total posture angle Γ _{j1 — n} and the total value Γ _{j2 — n} of posture angles with respect to the front-rear direction are expressed by the following equations (11-1) and (11-2), respectively.

  The constraint condition calculation unit 221 calculates the ratio of the total values of posture angles in different directions based on the measurement values of the same sensor device 110. Here, when the total value for one direction is relatively extremely small, there is a constraint that the change in the inclination of the back with respect to that direction is small, that is, the movement of the back with respect to that direction is restricted. It shows that. For this reason, when the total value for one direction is relatively extremely small, the constraint condition calculation unit 221 determines that the parameter for that direction does not contribute to the posture classification process.

Specifically, the constraint condition calculation unit 221 calculates a ratio (Γ _{j1 — n} / Γ _{j2 — n} ) and a ratio (Γ _{j2 — n} / Γ _{j1 — n} ) for each corresponding sensor device 110. When the ratio (Γ _{j1 — n} / Γ _{j2 — n} ) is less than or equal to a predetermined threshold value for all the sensor devices 110, the constraint condition calculation unit 221 does not contribute to the posture classification process by the posture angle corresponding to the left-right direction. Is determined. When the ratio (Γ _{j2 — n} / Γ _{j1 — n} ) is equal to or less than the same threshold value as described above for all the sensor devices 110, the constraint condition calculation unit 221 determines that the posture angle corresponding to the front-rear direction is the posture classification process. It is determined that it does not contribute. The threshold value can be arbitrarily set in the range of 0 to less than 1, and for example, a value such as “0.01” is set.

  Note that the determination condition may be modified as follows. For example, when the ratio corresponding to a predetermined number or more of sensor devices 110 is equal to or less than a threshold among the ratios calculated based on the posture angle corresponding to a certain direction, the constraint condition calculation unit 221 corresponds to that direction. It may be determined that the posture angle does not contribute to the posture classification process. As another method, the constraint condition calculation unit 221 further adds the total values of the posture angles based on the expressions (11-1) and (11-2) for a certain direction for all the sensor devices 110 and calculates them. The determination may be made by comparing the ratio of the total values that have been made with thresholds.

  If it is determined that there is a parameter corresponding to a direction that does not contribute to the posture classification process, the process of step S152 is executed. On the other hand, if it is determined that there is no parameter corresponding to a direction that does not contribute to the posture classification process, the process of step S111a is executed. Note that the processing in steps S111a and S112a executed in the latter case is the same processing as the posture parameter calculation processing in Processing Example 1 described above.

[Step S111a] If the constraint condition calculation unit 221 determines in step S151 that there is no parameter corresponding to a direction that does not contribute to posture classification, all the posture angles θ _{j1 — n} (N calculated by the feature amount calculation unit 212a are determined. ), Θ _{j2 — n} (N) is passed to the attitude parameter calculation unit 213a. The attitude parameter calculation unit 213a calculates a set of average vectors in the following procedure based on these attitude angles θ _{j1 — n} (N) and θ _{j2 — n} (N).

The posture angle Θ _j calculated for the j-th classification item is expressed as the above-described equation (5 ′). In addition, the average values Ave (θ _{j1 — n} ) and Ave (θ _{j2 — n} ) of the posture angles in the left-right direction and the front-rear direction in N measurements using the n-th sensor device 110 are the above-described equations (6), respectively. '-1) and (6'-2). Then, the average values Ave (θ _j1 ) and Ave (θ _j2 ) of the posture angles in the left-right direction and the front-rear direction in consideration of the measured values of all the sensor devices 110 are the expressions (6′-3) and (6 Represented by '-4). Therefore, the average vector Ave (Θ _j ) corresponding to the jth classification item is the average value of the posture angles represented by the equations (6′-3) and (6′-4) in the above equation (7). It is calculated by substituting Ave (θ _j1 ) and Ave (θ _j2 ).

[Step S112a] The posture parameter calculation unit 213a calculates one covariance matrix A _j based on the posture angles θ _{j1 — n} (N) and θ _{j2 — n} (N). The covariance matrix A _j includes the posture angle Θ _j calculated by the equation (5 ′) in step S111a and the average vector Ave (Θ _j ) calculated in step S111a using the posture angle Θ _j. It is calculated by substituting in the equation (8).

[Step S152] The constraint condition calculation unit 221a discards the parameter for the direction determined not to contribute to the posture classification process in Step S152.
Specifically, when it is determined that the posture angle θ _{j1 — n} (N) corresponding to the left-right direction does not contribute to the posture classification process, the constraint condition calculation unit 221a performs all posture angles θ corresponding to the left-right direction. _j1 _ discards _n a (n), and passes only the posture parameter calculating unit 213a attitude angle θ _j2 _ _n (n) corresponding to the front-rear direction. When it is determined that the posture angle θ _{j2 — n} (N) corresponding to the front-rear direction does not contribute to the posture classification process, the constraint condition calculation unit 221 a determines all posture angles θ _{j2 — n} corresponding to the front-rear direction. (N) is discarded and only the posture angle θ _{j1 — n} (N) corresponding to the left-right direction is passed to the posture parameter calculation unit 213a.

[Step S <b> 111 b] The posture parameter calculation unit 213 a calculates a set of average vectors using only the posture angle delivered from the constraint condition calculation unit 221.
The posture angle Θ _j calculated for the j-th classification item is represented as a matrix that does not include the posture angle discarded in step S152 in the above-described equation (5 ′). The posture angle Θ _j calculated in step S111a is a matrix of 2n rows and 2n columns, whereas the posture angle Θ _j calculated in step S111b is a matrix of n rows and n columns.

When it is determined in step S151 that the posture angle θ _{j1 — n} (N) corresponding to the left-right direction does not contribute to the posture classification process, the posture parameter calculation unit 213a performs the equations (6′-2) and (6 ′). -4), the average value Ave (θ _j2 ) corresponding to the front-rear direction is calculated. Then, the posture parameter calculation unit 213a calculates the average vector Ave (Θ _j ) by substituting only the average value Ave (θ _j2 ) corresponding to the front-rear direction into the above equation (7).

On the other hand, when it is determined in step S151 that the posture angle θ _{j2 — n} (N) corresponding to the front-rear direction does not contribute to the posture classification process, the posture parameter calculation unit 213a performs the equations (6′-1) and ( According to 6′-3), an average value Ave (θ _j1 ) corresponding to the left-right direction is calculated. Then, the posture parameter calculation unit 213a calculates the average vector Ave (Θ _j ) by substituting only the average value Ave (θ _j1 ) corresponding to the left and right direction into the above equation (7).

The average vector Ave (Θ _j ) calculated in step S111a is a 2n-dimensional vector, whereas the average vector Ave (Θ _j ) calculated in step S111b is an n-dimensional vector.

[Step S112b] The posture parameter calculation unit 213a calculates a set of covariance matrices A _j using only the posture angle delivered from the constraint condition calculation unit 221. The covariance matrix A _j includes the posture angle Θ _j calculated by the equation (5 ′) in step S111b and the average vector Ave (Θ _j ) calculated in step S111b using the posture angle Θ _j. It is calculated by substituting in the equation (8). The covariance matrix A _j calculated in step S112a is a 2n × 2n column matrix, while the covariance matrix A _j calculated in step S112b is an n × n matrix.

With the above procedure, the average vector Ave (Θ _j ) and the covariance matrix A _j corresponding to one classification item are calculated and stored in the posture parameter storage unit 214a.
Next, this process in Process Example 2 is executed in the same manner as in Process Example 1. That is, the feature quantity calculation unit 212a calculates n sets of posture angles θ _jN1 and θ _jN2 and outputs them to the posture classification processing unit 215a. The posture classification processing unit 215a calculates the probability P _j (θ) by substituting the posture angle θ represented by the above equation (10) into the above equation (9).

When the average vector Ave (Θ _j ) and the covariance matrix A _j corresponding to a certain classification item have been calculated by the processing in steps S111a and S112a in FIG. 20, the determination processing as to whether the subject's posture belongs to the classification item is This is performed by the same calculation as in Processing Example 1. In this case, the dimension d in Equation (9) is 2n.

On the other hand, if the average vector Ave (Θ _j ) and the covariance matrix A _j corresponding to a certain category item have been calculated by the processing in steps S111b and S112b in FIG. 20, it is determined whether the subject's posture belongs to that category item. The processing is performed using only the posture angle in one direction among the posture angles input from the feature amount calculation unit 212a. In this case, the dimension d in equation (9) is n.

  FIG. 21 is a graph showing an example of calculating the correct answer rate in Processing Example 2. FIG. 21 shows an example of the correct answer rate calculated by the correct answer rate calculating unit 216 when processing is performed with n = 2. FIG. 21 also shows combinations of mounting positions of two sensor devices 110, that is, combinations of sensor # 1 and sensor # 2, sensor # 1 and sensor # 3, and sensor # 2 and sensor # 3 shown in FIG. The correct answer rate is shown when processing is performed. Furthermore, in FIG. 21, for reference, the accuracy rate when processed by the posture classification apparatus 200 of the second embodiment using the sensors # 1, # 2, and # 3 alone (shown in FIG. 15). (Accuracy rate).

  As shown in FIG. 21, according to the processing example 2, when the sensor # 1 and the sensor # 2 are combined, and when the sensor # 2 and the sensor # 3 are combined, the corresponding sensor device 110 is displayed. A higher accuracy rate was obtained than when used alone. In comparison with Processing Example 1 shown in FIG. 18, a higher accuracy rate than Processing Example 1 was obtained for any combination of sensor devices 110.

<Processing example 3>
FIG. 22 is a diagram illustrating an example of the posture when the back is inclined obliquely.
In the above processing example 2, when the back is tilted in the front-rear direction or the left-right direction, a case is assumed in which the position variation of the back with respect to the direction orthogonal to the tilt direction is extremely small. However, there are few such cases in practice.

  For example, even when the subject tries to tilt his / her back in the front-rear direction, the back is often tilted in other directions besides the front-rear direction. FIG. 22 shows an example of such a case. For example, as shown on the left side of FIG. 22, when the back of the subject 140 is inclined diagonally right frontward, or as shown on the right side of FIG. In some cases, the back of the person leans forward diagonally to the left. Similarly, even when the subject tries to tilt his / her back in the left / right direction, the back is often tilted in other directions besides the left / right direction.

  As described above, even in a situation where the back is not tilted straight in the front-rear direction or the left-right direction, there is a direction with a small amount of fluctuation due to the constraint condition of the bone motion in the direction other than the front-rear direction or the left-right direction. it's possible. Such a parameter indicating the inclination of the back relative to the direction with a small amount of variation is considered to be a parameter that does not contribute to the posture classification process. For this reason, it is considered that the accuracy of posture classification can be improved by excluding the parameter indicating the inclination with respect to the direction from the posture parameter. However, such a direction differs depending on the subject, and it is difficult to uniquely determine as in the case assumed in Processing Example 1.

  In Processing Example 3 described below, the constraint condition calculation unit 221 estimates parameters that do not contribute to the posture classification processing by calculation, and the posture parameter correction unit 222 uses the posture parameter calculation unit 213a based on the estimation result. The calculated posture parameter is corrected. Specifically, the constraint condition calculation unit 221 decomposes the covariance matrix calculated by the posture parameter calculation unit 213a into eigenvalues and eigenvectors, and the eigenvalues corresponding to directions estimated not to contribute to the posture classification process and Determine the eigenvectors. The attitude parameter correction unit 222 recalculates the covariance matrix using the remaining eigenvalues and eigenvectors excluding the determined eigenvalues and eigenvectors among the eigenvalues and eigenvectors of the covariance matrix.

  Here, as an example, FIG. 23 and FIG. 24 show examples of posture angle distributions when the subject takes a posture corresponding to the classification item of “forward tilt-straight”, and the relationship between the posture angle distribution and the eigenvalues. Will be described.

  FIG. 23 is a diagram illustrating a distribution example of posture angles when a change in posture is limited to the front-rear direction. FIG. 24 is a diagram illustrating a distribution example of posture angles when the posture also changes in an oblique direction.

  An elliptical region 351 shown in FIG. 23 shows an image of a region in which the posture angle is distributed at a certain ratio when the change in the posture of the subject is substantially limited to the front-rear direction. In this case, among the eigenvalues of the covariance matrix calculated based on the attitude angle belonging to the elliptical region 351, the eigenvector direction D1 with respect to the maximum eigenvalue is the front-rear direction, that is, the left-right direction in FIG. On the other hand, in this case, since the amount of fluctuation in the horizontal direction of the back is very small, the eigenvector direction D2 for the extremely small eigenvalue among the eigenvalues of the covariance matrix is the horizontal direction, that is, the vertical direction in FIG. Become.

  On the other hand, an elliptical region 352 shown in FIG. 24 shows an image of a region in which the posture angle is distributed at a certain ratio when the posture of the subject fluctuates not only in the front-rear direction but also in the oblique direction. In this case, among the eigenvalues of the covariance matrix calculated based on the attitude angle belonging to the elliptical region 352, the eigenvector direction D3 with respect to the maximum eigenvalue is likely not to be the front-rear direction, that is, the left-right direction in FIG. . Then, the eigenvector direction D4 for an extremely small eigenvalue among the eigenvalues of the covariance matrix is likely not to be in the horizontal direction, that is, the vertical direction in FIG. 24 as in the example of FIG.

  In the processing example 3, the constraint condition calculation unit 221 calculates the cumulative contribution rate of the eigenvalues of the covariance matrix, thereby eliminating extremely small eigenvalues that do not contribute to the posture classification process from the eigenvalues of the covariance matrix.

  FIG. 25 is a flowchart illustrating an example of a processing procedure of posture parameter calculation in Processing Example 3. In Processing Example 3, the posture parameter calculation process (corresponding to step S12 in FIG. 7 and FIG. 9) is modified as follows.

[Step S111a] All posture angles θ _{j1 — n} (N) and θ _{j2 — n} (N) calculated by the feature amount calculation unit 212a are input to the posture parameter calculation unit 213a. The posture parameter calculation unit 213a calculates the average vector Ave (Θ _j ) corresponding to the j-th classification item in the same procedure as in step S111a in FIG. 20 (the same procedure as in processing example 1). The calculated average vector Ave (Θ _j ) is a 2n-dimensional vector.

[Step S112a] The posture parameter calculation unit 213a calculates one covariance matrix A _j corresponding to the j-th classification item in the same procedure as in step S112a in FIG. 20 (the same procedure as in Processing Example 1). The calculated covariance matrix A _j is a matrix of 2n rows and 2n columns.

[Step S161] constraint condition calculating section 221 performs eigenvalue decomposition for covariance matrix A _j calculated, and eigenvalue lambda _j, calculates the eigenvectors Q _j for the eigenvalues lambda _j.
Here, the covariance matrix A _j corresponding to the nth classification item can be expressed as A _j = Q _j Λ _j Q _j ^T. Further, the eigenvalue Λ _j is expressed as Λ _j = diag (λ ₁ ,..., Λ _d ), and the eigenvector Q _j is expressed as Q _j = (q ₁ ,..., Q _d ). Λ ₁ ,..., Λ _d are in descending order.

[Step S162] The constraint condition calculation unit 221 sets the variable b to “1”.
[Step S163] The constraint condition calculation unit 221 calculates the cumulative contribution rate ξ (b) according to the following equation (13).

[Step S164] constraint condition calculating section 221, the cumulative contribution rate calculated xi] (b) it is determined whether there are more threshold xi] _c. The threshold value ξ _c is a value determined empirically, and is desirably as large as possible (for example, 0.999) within a range that satisfies the condition of 0 <ξ _c <1.

If the cumulative contribution rate ξ (b) is less than the threshold value ξ _c , the process of step S165 is executed. If the cumulative contribution rate ξ (b) is greater than or equal to the threshold value ξ _c , the process of step S166 is executed.

  [Step S165] The constraint condition calculation unit 221 increments the variable b by “1”. Thereafter, the calculation in step S163 is performed again using the incremented variable b.

The variable b when the cumulative contribution ratio xi] (b) is determined to be equal to or greater than the threshold value xi] _c at Step S166] Step S164, the B_c. The constraint condition calculation unit 221 determines eigenvalues and eigenvectors used for recalculation of the covariance matrix as shown in the following equations (14) and (15), respectively.
Λ ′ _j = (λ ₁ ,..., Λ _{b — c} ) (14)
Q ′ _j = (q ₁ ,..., Q _{b — c} ) (15)
The attitude parameter correction unit 222 recalculates the covariance matrix A ′ _{j according} to the following equation (8 ′) based on the equations (14) and (15).
A ′ _j = Q ′ _j Λ ′ _j Q ′ _j ^T (8 ′)
The posture parameter correction unit 222 uses the average vector Ave (Θ _j ) calculated in step S111a and the covariance matrix A ′ _j recalculated in step S166 as posture parameters corresponding to the nth classification item. Stored in the posture parameter storage unit 214a.

Next, this process in Process Example 2 is executed in the same manner as in Process Example 1. That is, the feature quantity calculation unit 212a calculates n sets of posture angles θ _jN1 and θ _jN2 and outputs them to the posture classification processing unit 215a. The posture classification processing unit 215a refers to the posture parameter stored in the posture parameter storage unit 214a, and substitutes the posture angle θ expressed by the above-described equation (10) into the above-described equation (9), thereby generating a probability. P _j (θ) is calculated. Then, the posture classification processing unit 215a determines and outputs the classification item corresponding to the probability having the highest value among the calculated n probabilities as the classification item to which the posture of the subject belongs.

The dimension d of Equation (9) is 2n at the maximum. However, in the process of FIG. 25, among the eigenvalues λ ₁ ,..., Λ _d of the covariance matrix A _j , the dimensions are equal to the number of eigenvalues that are not used in the calculation of step S166 (ie, 2n−b_c). d decreases.

Here, in the process at the step S162~S164 the minimum value of the variable b as the cumulative contribution ratio xi] (b) does not exceed the threshold ξ _c (= b_c) is calculated. Then, from the eigenvalues (λ ₁ ,..., Λ _d ), the eigenvalues from the head to the b_c-th eigenvalue are extracted, and the other eigenvalues, that is, the extreme values that do not contribute to the posture classification process are extracted. Eigenvalues that are too small are excluded from the target of the covariance matrix calculation in step S166.

In other words, from among the eigenvalues and eigenvectors of the covariance matrix A _j , which is one of the parameters used for the posture classification process, the direction of the back tilt caused by the constraint condition of the bone motion is extremely small. Corresponding eigenvalues and eigenvectors are determined. Then, the covariance matrix is recalculated using the parameters excluding the determined eigenvalues and eigenvectors.

Thereby, even when a direction in which the back tilt is small cannot be predicted, the posture classification accuracy can be improved.
FIG. 26 is a graph illustrating an example of calculating the correct answer rate in Processing Example 3. FIG. 26 shows an example of the correct answer rate calculated by the correct answer rate calculation unit 216 when the process is performed with n = 2. In FIG. 26, as combinations of mounting positions of the two sensor devices 110, combinations of sensor # 1 and sensor # 2, sensor # 1 and sensor # 3, and sensor # 2 and sensor # 3 shown in FIG. The correct answer rate is shown when processing is performed. Further, in FIG. 26, for reference, the accuracy rate when the sensors # 1, # 2, and # 3 are used alone and processed by the posture classification apparatus 200 of the second embodiment (shown in FIG. 15). (Accuracy rate).

  As shown in FIG. 26, according to Processing Example 3, a higher accuracy rate was obtained with any combination of sensor devices 110 than when the corresponding sensor device 110 was used alone. Moreover, even when compared with the processing example 1 shown in FIG. 18 and the processing example 2 shown in FIG. 21, a high accuracy rate was obtained with any combination of the sensor devices 110.

  FIG. 27 is a graph illustrating an example of calculating the correct answer rate for each classification item. In the accuracy rate list 400 shown in FIG. 27, each of the processing example 1, the processing example 2, and the processing example 3 in the second embodiment and the third embodiment is calculated for each classification item. The correct answer rate is shown. The classification items are represented by the identification numbers shown in FIG. In addition, the numerical values shown in the item of average value are those shown in the graphs of FIGS.

  As shown in FIG. 27, according to the processing of the second embodiment, particularly when sensor # 1 is used, the classification processing has succeeded to some extent for all 13 types of classification items. Moreover, it turns out that the classification accuracy of 13 types of classification items improves as a whole in order of processing examples 1, 2, and 3. In particular, the combination of sensors # 1 and # 2 in process example 2, the combination of sensors # 2 and # 3 in process example 2, and the combination of all sensor devices 110 in process example 3 are 60 for all 13 types of classification items. % Accuracy rate was obtained. Further, the combination of sensors # 1 and # 3 in Processing Example 3 and the combination of sensors # 2 and # 3 in Processing Example 3 gave an accuracy rate of 80% or more for all 13 types of classification items. As described above, according to each of the above-described embodiments, it is possible to classify postures not only according to the back-and-forth and left-right inclinations but also according to the curved state of the back.

<Processing example 4>
The process example 4 is a combination of the process of the process example 2 and the process of the process example 3.
FIG. 28 is a flowchart illustrating an example of a processing procedure of posture parameter calculation in Processing Example 4. In the process example 4, the posture parameter calculation process (corresponding to step S12 in FIG. 7 and FIG. 9) is modified as follows. In FIG. 28, processing steps in which the same processing as in FIGS. 20 and 25 is performed are denoted by the same reference numerals, and description thereof is omitted.

  First, the constraint condition calculation unit 221 determines whether there is a parameter corresponding to a direction that does not contribute to the posture classification process based on the posture angle calculated by the feature amount calculation unit 212a, as in the case of the processing example 2. (Step S151). When there is a parameter corresponding to a direction that does not contribute to the posture classification process, the parameter (posture angle) corresponding to the direction is discarded, and the posture parameter calculation unit 213a only sets the parameter (posture angle) corresponding to the remaining direction. The average vector and the covariance matrix are calculated using them (steps S152, S111b, S111b).

  On the other hand, if it is not determined in step S151 that there is a parameter in a direction that does not contribute to the posture classification process, the same process as in process example 3 is executed. That is, the mean vector and the covariance matrix are calculated by the posture parameter calculation unit 213a in the same procedure as in the processing example 1 (steps S111a and S112a). The constraint condition calculation unit 221 excludes eigenvalues that do not contribute to the attitude classification process from the eigenvalues of the covariance matrix based on the calculation of the cumulative contribution rate. Then, posture parameter correction section 222 recalculates the covariance matrix using the remaining eigenvalues other than the excluded eigenvalues and eigenvectors for these eigenvalues (steps S161 to S166).

[Fourth Embodiment]
In the second embodiment and the third embodiment described above, the measurement value measured by the sensor device 110 is transmitted to the posture classification devices 200 and 200a via the terminal devices 120 and 120a, and the posture classification device 200. , 200a, posture parameter calculation and posture classification processing are performed. In contrast, in the fourth embodiment, the terminal device is configured to perform posture parameter calculation and posture classification processing.

  FIG. 29 is a block diagram illustrating a configuration example of a terminal device according to the fourth embodiment. A plurality of sensor devices 110 are connected to the terminal device 120b shown in FIG. Further, the terminal device 120b has a processing function similar to that of the posture classification device 200a illustrated in FIG. However, the data acquisition unit 211a directly acquires the measurement value measured by the sensor device 110 from the sensor device 110.

  In FIG. 29, the processing function of the posture classification device 200a in the third embodiment is provided in the terminal device 120b. However, the processing function of the posture classification device 200 in the second embodiment is provided in the terminal device 120b. Also good.

  In the second to fourth embodiments, the learning process and the main process are performed by one apparatus. However, these processes may be performed by different apparatuses. In this case, the posture parameter calculated by the device on which the learning process has been performed is input to the device on which this process is performed via a portable recording medium or a network.

  In addition, the processing functions of the devices (the information processing device 1, the posture classification devices 200 and 200b, and the terminal devices 120, 120a, and 120b) described in the above embodiments can be realized by a computer. In that case, a program describing the processing contents of the functions that each device should have is provided, and the processing functions are realized on the computer by executing the program on the computer. The program describing the processing contents can be recorded on a computer-readable recording medium. Examples of the computer-readable recording medium include a magnetic storage device, an optical disk, a magneto-optical recording medium, and a semiconductor memory. Examples of the magnetic storage device include a hard disk device (HDD), a flexible disk (FD), and a magnetic tape. Optical disks include DVD (Digital Versatile Disc), DVD-RAM, CD-ROM (Compact Disc-Read Only Memory), CD-R (Recordable) / RW (ReWritable), and the like. Magneto-optical recording media include MO (Magneto-Optical disk).

  When distributing the program, for example, a portable recording medium such as a DVD or a CD-ROM in which the program is recorded is sold. It is also possible to store the program in a storage device of a server computer and transfer the program from the server computer to another computer via a network.

  The computer that executes the program stores, for example, the program recorded on the portable recording medium or the program transferred from the server computer in its own storage device. Then, the computer reads the program from its own storage device and executes processing according to the program. The computer can also read the program directly from the portable recording medium and execute processing according to the program. In addition, each time a program is transferred from a server computer connected via a network, the computer can sequentially execute processing according to the received program.

DESCRIPTION OF SYMBOLS 1 Information processing apparatus 2 Measurement value acquisition part 3 Parameter calculation part 4 Classification processing part 5 Memory | storage device 11 Test subject 12 Sensor 21 Parameter 22 Probability model 31 1st measurement value 32 2nd measurement value

Claims (12)

For each posture type that classifies the posture of the human body into a plurality of types, obtain a plurality of first measurement values obtained by measuring the inclination of the human body with respect to a plurality of directions,
Based on the plurality of first measurement values for each posture type, a parameter for generating a probability model of measurement values is calculated for each posture type, and stored in a storage device;
Obtaining a second measured value obtained by measuring the inclination of the human body with respect to the same plurality of directions as the first measured value,
Based on the second measurement value and the probability model for each posture type respectively generated based on the parameters stored in the storage device, the posture of the human body to be measured by the second measurement value is determined. Classify into one of a plurality of types of postures,
A posture classification method characterized by that.   2. The posture classification method according to claim 1, wherein the probability model for each posture type is a probability model of measurement values according to a multidimensional normal distribution. The parameters calculated for each posture type are an average vector and a covariance matrix based on the first measurement value for each measured tilt direction,
In the process of classifying the posture, the probability is calculated for each posture type based on the second measurement value, the average vector and the covariance matrix for each posture type, and the posture of the human body to be measured is calculated as follows: Classify into the posture types with the highest calculated probability,
The posture classification method according to claim 2, wherein: In the acquisition of the first measurement value, the measurement value measured by a plurality of sensors each measuring the inclination of the human body with respect to a plurality of directions is acquired,
In the acquisition of the second measurement value, the inclination of the human body with respect to a plurality of directions is measured, and the measurement values measured by the same number of sensors as in the measurement of the first measurement value are acquired.
The posture classification method according to claim 1. Based on the plurality of first measurement values for each posture type, a restriction direction in which the inclination of the human body to be measured is limited among the plurality of directions indicated by the first measurement value is determined for each posture type. ,
Further comprising processing,
In the calculation of the parameter for each posture type, when the restriction direction is determined for one posture type among the plurality of posture types, the first measurement value corresponding to the one posture type is calculated. , Using the remaining first measurement values excluding the first measurement value indicating the inclination with respect to the restriction direction, a parameter corresponding to the one posture type is calculated and stored in the storage device.
The posture classification method according to claim 4. Based on the parameters calculated for each posture type, for each posture type, determining a restriction direction in which the inclination of the human body to be measured is limited,
When the restriction direction is determined for one of the plurality of posture types, the parameter indicating the inclination with respect to the determined restriction direction is excluded so that the parameter corresponding to the one posture type is excluded. Recalculate the parameters and store them in the storage device;
The posture classification method according to claim 4, further comprising a process. The probability model for each posture type is a probability model of a measurement value according to a multidimensional normal distribution,
The parameters calculated for each posture type are an average vector and a covariance matrix based on the first measurement value for each measured tilt direction,
In the determination of the restriction direction, based on the cumulative contribution ratio of the eigenvalues of the covariance matrix calculated for each posture type, the eigenvalue to be processed is selected for each posture type from among the eigenvalues of the covariance matrix,
In the recalculation of the parameter, when an eigenvalue to be processed is selected for the one posture type, a covariance corresponding to the one posture type is used by using the selected eigenvalue and an eigenvector for the eigenvalue. Recalculate the matrix,
The posture classification method according to claim 6.   The plurality of sensors for measuring the plurality of first measurement values and the plurality of the second measurement values are respectively mounted on the back of the human body at different positions with respect to the direction along the spine. The posture classification method according to claim 1, wherein: A measurement value acquisition unit for acquiring a measurement value obtained by measuring the inclination of the human body with respect to a plurality of directions;
When the measurement value acquisition unit acquires a plurality of first measurement values obtained by measuring the inclination of the human body with respect to a plurality of directions for each posture type obtained by classifying the postures of the human body into a plurality of types, a plurality of the first measurement values for each posture type are obtained. A parameter calculation unit for calculating a parameter for generating a probability model of the measurement value for each posture type based on the measurement value of 1 and storing the parameter in the storage device;
An information processing apparatus comprising:   When the measurement value acquisition unit acquires a second measurement value obtained by measuring the inclination of the human body with respect to the same plurality of directions as the first measurement value, the second measurement value and a parameter stored in the storage device A classification processing unit that classifies the posture of the human body to be measured by the second measurement value into one of a plurality of types of postures based on the probability model for each posture type generated based on The information processing apparatus according to claim 9, further comprising: On the computer,
For each posture type that classifies the posture of the human body into a plurality of types, obtain a plurality of first measurement values obtained by measuring the inclination of the human body with respect to a plurality of directions,
Based on the plurality of first measurement values for each posture type, a parameter for generating a probability model of the measurement value is calculated for each posture type and stored in a storage device.
A program characterized by causing processing to be executed. In the computer,
Obtaining a second measured value obtained by measuring the inclination of the human body with respect to the same plurality of directions as the first measured value,
Based on the second measurement value and the probability model for each posture type respectively generated based on the parameters stored in the storage device, the posture of the human body to be measured by the second measurement value is determined. Classify into one of a plurality of types of postures,
The program according to claim 11, further executing a process.

Images