JP6834180B2 - Step number learning method, step number learning program, information processing device, walking data processing method, walking data processing program and walking data processing device - Google Patents

Description

The present invention relates to a step number learning method, a step number learning program, and an information processing device.

A pedometer that is installed in a mobile terminal such as a smartphone and measures the number of steps of a user is used. For example, a pedometer measuring system acquires 3-axis acceleration data from a 3-axis acceleration sensor mounted on a smartphone, generates a one-dimensional waveform from the three-axis acceleration data, and uses the generated one-dimensional waveform. Count the number of steps.

Japanese Unexamined Patent Publication No. 2003-302296 JP-A-2009-223744 Japanese Unexamined Patent Publication No. 2008-171347 JP-A-2015-135662

However, in the above technique, since the parameters of the built-in algorithm are adjusted before the product is shipped, the step count measurement using the same parameters is performed for any user. Therefore, it cannot be said that the accuracy of the measured number of steps is high. In addition, it is difficult to adjust the parameters after shipping the product because the individual gaits are different.

In one aspect, it is an object of the present invention to provide a pedometer learning method, a pedometer learning program, and an information processing device capable of improving the accuracy of pedometer measurement.

In the first plan, the computer executes a process of acquiring the timing of landing or taking off for each step of the user and a process of acquiring acceleration data indicating a time-series change. The computer executes a process of extracting acceleration data corresponding to the period of one step of the user from the acquired acceleration data by using the acquired landing or takeoff timing. The computer executes a process of setting a one-dimensional ideal waveform as a determination criterion for counting the number of steps of the user from the extracted acceleration data.

According to one embodiment, the accuracy of step count measurement can be improved.

FIG. 1 is a diagram illustrating a mobile terminal according to the first embodiment. FIG. 2 is a diagram showing a hardware configuration example of the mobile terminal according to the first embodiment. FIG. 3 is a functional block diagram showing a functional configuration of the mobile terminal according to the first embodiment. FIG. 4 is a diagram illustrating the measurement of stride time. FIG. 5 is a diagram illustrating sampling and fitting of an ideal output waveform. FIG. 6 is a diagram illustrating input discrete vector data. FIG. 7 is a diagram illustrating fitting of input discrete vector data. FIG. 8 is a diagram illustrating the relationship between input and output. FIG. 9 is a diagram illustrating processing at the time of step count measurement. FIG. 10 is a diagram illustrating step count measurement. FIG. 11 is a flowchart showing the flow of processing.

Hereinafter, examples of the step number learning method, the step number learning program, and the information processing apparatus disclosed in the present application will be described in detail with reference to the drawings. The present invention is not limited to this embodiment.

[Explanation of mobile terminal 10]
FIG. 1 is a diagram illustrating a mobile terminal 10 according to the first embodiment. The mobile terminal 10 is an information processing device such as a smartphone. The mobile terminal 10 is equipped with a step count measuring system, and measures the number of steps of a user by using acceleration data of three axes.

As shown in FIG. 1, the pedometer measuring system learns parameters according to the user's gait before measuring the number of steps, and uses the learned parameters to count so-called zero cross points to measure the number of steps of the user. .. For example, a pedometer measuring system inputs acceleration data of three axes, outputs an ideal waveform (for example, a cos waveform) representing the number of steps, and counts the number of steps.

More specifically, the pedometer measurement system performs supervised learning using the acceleration data of the three axes and the ideal waveform representing the number of steps as supervised data, and trains each parameter of the learning pedometer system. After learning, the pedometer measuring system counts the number of steps according to the waveform output using the learning result when the acceleration data of the three axes is input.

Therefore, even after the product of the mobile terminal 10 equipped with the step count measurement system is shipped, the step count of the user can be measured after adjusting the parameters according to the gait of the user. The accuracy can be improved.

[Hardware configuration of mobile terminal 10]
Next, the hardware configuration of the mobile terminal 10 equipped with the step count measurement system will be described. FIG. 2 is a diagram showing a hardware configuration example of the mobile terminal 10 according to the first embodiment. As shown in FIG. 2, the mobile terminal 10 includes a wireless unit 1, an acceleration sensor 2, an audio input / output unit 3, a storage device 4, and a display device 5. The hardware shown here is an example, and may have other hardware such as a GPS (Global Positioning System) receiver.

The wireless unit 1 transmits / receives data to / from another terminal via the antenna 1a. The acceleration sensor 2 is a sensor that measures the acceleration of the mobile terminal 10 and outputs it to the processor 6, and is a three-axis acceleration sensor that measures acceleration in three directions of the xyz axis.

The audio input / output unit 3 collects sound through a microphone and outputs sound through a speaker. The storage device 4 is a storage device for storing programs and data, such as a memory and a hard disk. The display device 5 is a display, a touch panel, or the like that displays various information.

The processor 6 is a CPU (Central Processing Unit) or the like, reads a program in which the contents of various processes described later are defined from the storage device 4 and executes the program, and executes various processes related to the step count, that is, the step count system.

[Functional configuration]
FIG. 3 is a functional block diagram showing a functional configuration of the mobile terminal 10 according to the first embodiment. For example, the mobile terminal 10 executes the step count measurement system by executing each processing unit shown in FIG. Each functional unit described with reference to FIG. 3 is an example of an electronic circuit included in the processor 6 and an example of a process executed by the processor 6.

As shown in FIG. 3, the mobile terminal 10 includes an acceleration acquisition unit 11, a calculation unit 12, a data length determination unit 13, an ideal waveform determination unit 14, a sampling unit 15, a data division unit 16, an adjustment unit 17, and a learning unit 18. , A model generation unit 19, and a measurement unit 20.

The acceleration acquisition unit 11 is a processing unit that acquires acceleration data from the three-axis acceleration sensor 2. For example, the acceleration acquisition unit 11 acquires acceleration data at the time of parameter learning or step count measurement and outputs it to the data division unit 16 or the calculation unit 12.

The calculation unit 12 is a processing unit that calculates the stride time. Specifically, the calculation unit 12 acquires acceleration data from the acceleration acquisition unit 11, while acquiring the timing at which the foot lands or takes off every two steps after the user starts walking. As a method of acquiring the timing of landing or taking off, the user can also grasp the switch and operate the switch at the timing of landing or taking off, attach a sensor to the user's foot or shoes, and acquire by the sensor. You can also do it.

FIG. 4 is a diagram illustrating the measurement of stride time. As shown in FIG. 4, the calculation unit 12 acquires acceleration data for each of the x-axis, y-axis, and z-axis, while acquiring the timing (switch on) when the user presses the switch. Then, the calculation unit 12 calculates the stride time (m _{j + 1} −m _j ) from the timing when the switch is pressed, and outputs it to the data length determination unit 13. Here, j is 1, 2, ..., R.

The data length determination unit 13 is a processing unit that determines the data length (2n + 1) using the stride time calculated by the calculation unit 12. For example, the data length determination unit 13 defines the data length (2n + 1) as _{the first odd number of max (m j + 1} −m _j ; j = 1, 2, ..., R) or more. Then, the data length determination unit 13 outputs the determined data length (2n + 1) to the ideal waveform determination unit 14.

The ideal waveform determination unit 14 is a processing unit that determines an output ideal waveform, which is an ideal waveform output when the acceleration data of the three axes is used as input data. For example, the ideal waveform determination unit 14 is represented by the equation (1) using the data length (2n + 1) determined by the data length determination unit 13 and the stride time (m _{j + 1} −m _{j) calculated by the calculation unit 12.} The output ideal waveform _pj (t) is determined. This output ideal waveform _pj (t) has a feature of forming a waveform with two cycles in stride time, and r waveforms are created. In addition, "A" in equation (1) is a positive constant, and "t" is time.

The sampling unit 15 is a processing unit that generates a fitting curve by using a curve fitting technique for _{the output ideal waveform pj (t).} FIG. 5 is a diagram illustrating sampling and fitting of an ideal output waveform. For example, the sampling unit 15 obtains discrete vector data by sampling as shown in FIG. 5 in order to match _{the output ideal waveform pj (t) with the data length. Then, the sampling unit 15 creates a fitting curve s j} (t) of the discrete vector data by curve fitting using the discrete Fourier transform (DFT), and outputs the fitting curve s j (t) to the learning unit 18. A total of r fitting curves s _j (t) are also created. As the curve fitting technique, the technique described in JP-A-2015-135662 can be adopted.

The data division unit 16 is a processing unit that divides the three-dimensional acceleration data acquired by the acceleration acquisition unit 11. FIG. 6 is a diagram illustrating input discrete vector data. As shown in FIG. 6, the data division unit 16 divides the three-dimensional acceleration data acquired by the acceleration acquisition unit 11 into data lengths of “2n + 1”, and inputs x-axis, y-axis, and z-axis discrete vectors. Create data. Here, a total of r input discrete vector data on the x-axis, y-axis, and z-axis are created.

The adjustment unit 17 is a processing unit that creates input discrete vector data using acceleration data and the number of reference samples when walking two steps, and creates a fitting curve of input discrete vector data using curve fitting technology. .. FIG. 7 is a diagram illustrating fitting of input discrete vector data. As shown in FIG. 7, the adjusting unit 17 performs curve fitting using the discrete Fourier transform (DFT) on the input discrete vector data (see FIG. 6) obtained by the data dividing unit 16 to obtain the input discrete vector. A fitting curve is created and output to the learning unit 18. Each of the fitting curves x _j (t), y _j (t), and z _j (t) is also created in r pieces.

The learning unit 18 is a processing unit that learns parameters for obtaining an ideal output waveform. FIG. 8 is a diagram illustrating the relationship between input and output. As shown in FIG. 8, the learning unit 18 is generated by the sampling unit 15 by inputting _{r fitting curves x j} (t), y _j (t), and z _{j (t) generated by the adjusting unit 17.} The parameters for obtaining the r fitting curves s _j (t) are learned.

For example, the learning unit 18 has _{the relationship shown in FIG. 8 because the fitting curves x j} (t), y _j (t), z _j (t), and s _j (t) are all finite Fourier series. The mathematical model to have can be expressed by the equation (2) that compares the input and output Fourier series. The a on the right side of the mathematical model (Equation (2)) represents the inner product, and is derived from the property that addition and multiplication of Fourier series are also Fourier series. _{Since D v,} v'on the right side of the mathematical model (Equation (2)) is unknown, the learning unit 18 has r x _j (t), y _j (t), z _j (t) as teacher data. ), S _j (t) is used to learn and determine the coefficients of _{D v, v'.}

For example, the learning unit 18 rewrites the equation (2) as the equation (3). Equation (3) is an equation obtained by extracting the same exponential portion of the mathematical model (Equation (2)), and is a simultaneous equation with _{D v and v'as unknowns. Therefore, the solution of D v, v'of} this simultaneous equation can be uniquely determined by using linear regression such as the least squares method. Then, the learning unit 18 outputs the uniquely determined _{solutions of D v, v'as} parameters to the model generation unit 19.

Since x _j (t), y _j (t), z _j (t), and s _j (t) are Fourier series, it is as simple as equation (3) from the equation including the function of equation (2). However, if it is not a Fourier series, such a conversion is not easy and the amount of calculation for finding a solution increases, so it is not a preferable method.

The model generation unit 19 is a processing unit that outputs a waveform used for step count measurement by using the acceleration data and the learning result by the learning unit 18. FIG. 9 is a diagram illustrating processing at the time of step count measurement. As shown in FIG. 9, the model generation unit 19 slides the 3-axis acceleration data output from the 3-axis acceleration sensor 2 in a window having a window width of "2n + 1" to obtain a portion having a data length of "2n + 1". Acceleration data is created sequentially.

Then, the model generation unit 19 performs the curve fitting on the created partial acceleration data to create fitting curves x _j (t), y _j (t), and z _j (t). Next, the model generation unit 19 considers a calculation model that outputs the _{curve u j (t) by inputting the obtained fitting curves x j} (t), y _j (t), and z _j (t). Then, the curve u _j (t) can be expressed by the equation (4) _{by using x j} (t), y _j (t), z _j (t) and D _{v, v'obtained by learning.} it can. Then, the model generation unit 19 outputs the waveform represented by the equation (4) to the measurement unit 20.

The measurement unit 20 is a processing unit that measures the number of steps of the user based on the waveform generated by the model generation unit 19. Specifically, the measuring unit 20 measures the number of steps by counting so-called zero crosses.

FIG. 10 is a diagram illustrating step count measurement. The horizontal axis of the graph in FIG. 10 is a time and the vertical axis represents wavelength _(u j _(t)). When the waveform output from the model generation unit 19 is the waveform shown in FIG. 10, the measurement unit 20 counts the black points at which the value of the waveform becomes approximately 0. Then, the measuring unit 20 _counts the walked 0.5 steps at the point where the _u j (t) ≒ 0. That is, the measuring unit 20 measures the number of steps of the user with 0.5 steps between the black dots.

_{In addition, when the output waveform of uji} (t) has a shape far from the output ideal waveform or is not similar to the output waveform, the measurement unit 20 has an amplitude even if the waveform is close to the periodic waveform such as the output ideal waveform. When is small (for example, when it is equal to or less than a predetermined value), it is determined that the user is not in a walking state. That is, the measuring unit _20, when the waveform close to the waveform output ideal waveform of _u j (t), is determined to be a walking _state, at a point t which is a _{u j (t) ≒ 0,} 0.5 cheek Count as being.

[Processing flow]
FIG. 11 is a flowchart showing the flow of processing. The process related to pedometer includes a learning process phase and a pedometer execution process phase.

As shown in FIG. 11, the acceleration acquisition unit 11 acquires acceleration data (S101), and the calculation unit 12 acquires the timing at which the foot lands or takes off every two steps after the user starts walking. Then, the stride time of the user is calculated (S102).

Subsequently, the data length determination unit 13 determines the data length (2n + 1) using the calculated stride time (S103), and the ideal waveform determination unit 14 outputs the acceleration data of the three axes as input data. The output ideal waveform, which is the ideal waveform to be formed, is determined (S104).

After that, the sampling unit 15 samples the output ideal waveform to acquire the discrete vector data (S105), and creates (extracts) the fitting curve of the discrete vector data by curve fitting using the discrete Fourier transform (DFT) (S105). S106).

Subsequently, the data division unit 16 divides the three-dimensional acceleration data acquired by the acceleration acquisition unit 11 with a data length of “2n + 1” (S107). Further, the adjustment unit 17 creates input discrete vector data using the acceleration data when walking two steps and the number of reference samples, performs curve fitting (S108), and creates a fitting curve of the input discrete vector data (S108). S109).

Then, the learning unit 18 learns the parameters of the mathematical model for obtaining _{the fitting curve s j} (t) generated by using the output ideal waveform from the fitting curve generated from the acceleration data of the three axes (S110). ..

When the acceleration data is measured after the learning is completed (S111: Yes), the model generation unit 19 divides the acceleration data (S112), executes curve fitting (S113), and extracts the fitting curve (S114). .. Then, the model generation unit 19 generates an output curve using the extracted fitting curve and the learning result (S115). After that, the measuring unit 20 counts the number of steps according to the waveform of the output curve (S116).

[effect]
As described above, the mobile terminal 10 can adjust each parameter of the pedometer measurement system after the product is shipped, and it becomes possible to create a pedometer system customized for each individual. Therefore, since the mobile terminal 10 can measure the number of steps of the user after adjusting the parameters according to the gait of the user, the accuracy of the number of steps measurement can be improved.

Further, since the mobile terminal 10 can accurately adjust the parameters according to the gait of the user, the amplitude is even when the output waveform has a shape far from the waveform of two cycles or a waveform close to two cycles. When is small, it can be determined that the user is not in a walking state. As a result, the mobile terminal 10 can accurately determine the walking state and the non-walking state.

Although the examples of the present invention have been described so far, the present invention may be implemented in various different forms other than the above-described examples.

[Applicable target]
In the above embodiment, the mobile terminal 10 has been described as an example, but the present invention is not limited to this, and each function (walking measurement system) shown in FIG. 3 is incorporated into a product such as ubiquitous wear and used. can do.

[Curve fitting]
In the above embodiment, an example of adopting the technique described in Japanese Patent Application Laid-Open No. 2015-135662 has been described, but the present invention is not limited thereto. For _example, when an example _x j _(t), plotted x _{1 (t),} x 2 a _{(t), ··· x r (} t), the horizontal axis time t and the vertical axis as _x j (t) Any technique can be adopted as long as it is a technique that generates a curve that passes through each point.

[system]
Further, each configuration of each device shown in FIG. 3 does not necessarily have to be physically configured as shown in the drawing. That is, it can be dispersed or integrated in any unit. For example, the model generation unit 19 and the measurement unit 20 can be integrated. Further, each processing function performed by each device may be realized by a CPU and a program analyzed and executed by the CPU, or may be realized as hardware by wired logic.

Further, among the processes described in this embodiment, all or a part of the processes described as being automatically performed can be manually performed. Alternatively, all or part of the processing described as being performed manually can be automatically performed by a known method. In addition, the processing procedure, control procedure, specific name, and information including various data and parameters shown in the above document and drawings can be arbitrarily changed unless otherwise specified.

10 Mobile terminal 11 Acceleration acquisition unit 12 Calculation unit 13 Data length determination unit 14 Ideal waveform determination unit 15 Sampling unit 16 Data division unit 17 Adjustment unit 18 Learning unit 19 Model generation unit 20 Measurement unit

Claims (9)

The computer
Get the timing of landing or taking off for each step of the user,
Acquire acceleration data showing changes over time,
Using the acquired landing or takeoff timing, the acceleration data corresponding to the one-step period of the user is extracted from the acquired acceleration data.
From the extracted acceleration data, a one-dimensional ideal waveform that serves as a criterion for counting the number of steps of the user is set.
Curve fitting is performed on the discrete vector data generated from the ideal waveform to calculate the first fitting curve.
A second fitting curve is calculated by performing curve fitting on the input discrete vector data generated using the acceleration data when the user walks two steps.
Input the second fitting curve and learn the parameters satisfying the numerical model from which the first fitting curve is output.
A step count learning method characterized by executing a process. In the process to be set, the acceleration data and the stride time when the user walks two steps are acquired, the data length of the acceleration data to be input is determined according to the stride time, and the determined data length is used. The step number learning method according to claim 1, wherein the ideal waveform is set. When the acceleration data is acquired after learning the parameters, the acquired acceleration data is divided by the data length to generate divided acceleration data, and the output waveform is generated by using the divided acceleration data and the parameters. The process to generate and
The step number learning method according to claim 2, wherein the computer executes a process of counting the number of steps based on the output waveform. The counting process is characterized in that when the output waveform has an amplitude smaller than a predetermined amplitude or is dissimilar to the ideal waveform, it is determined that the waveform is not in a walking state and the counting of the number of steps is suppressed. The step count learning method according to claim 3. On the computer
Get the timing of landing or taking off for each step of the user,
Acquire acceleration data showing changes over time,
Using the acquired landing or takeoff timing, the acceleration data corresponding to the one-step period of the user is extracted from the acquired acceleration data.
From the extracted acceleration data, a one-dimensional ideal waveform that serves as a criterion for counting the number of steps of the user is set.
Curve fitting is performed on the discrete vector data generated from the ideal waveform to calculate the first fitting curve.
A second fitting curve is calculated by performing curve fitting on the input discrete vector data generated using the acceleration data when the user walks two steps.
Input the second fitting curve and learn the parameters satisfying the numerical model from which the first fitting curve is output.
A step count learning program characterized by executing processing. An acquisition unit that acquires the timing of landing or taking off for each step of the user,
An acquisition unit that acquires acceleration data indicating changes over time,
An extraction unit that extracts acceleration data corresponding to the period of one step of the user from the acquired acceleration data by using the acquired landing or takeoff timing.
From the extracted acceleration data, a setting unit that sets a one-dimensional ideal waveform that serves as a judgment criterion for counting the number of steps of the user, and a setting unit.
A calculation unit that performs curve fitting on the discrete vector data generated from the ideal waveform to calculate the first fitting curve, and
A calculation unit that performs curve fitting on the input discrete vector data generated using the acceleration data when the user walks two steps and calculates a second fitting curve.
A learning unit that inputs a second fitting curve and learns parameters that satisfy a numerical model from which the first fitting curve is output.
An information processing device characterized by having. The computer
Get the timing of landing or taking off for each step of the user,
Acquire acceleration data showing changes over time,
Using the acquired landing or takeoff timing, the acceleration data corresponding to the one-step period of the user is extracted from the acquired acceleration data.
From the extracted acceleration data, a one-dimensional ideal waveform that serves as a criterion for counting the number of steps of the user is set.
A walking data processing method characterized by executing processing. On the computer
Get the timing of landing or taking off for each step of the user,
Acquire acceleration data showing changes over time,
Using the acquired landing or takeoff timing, the acceleration data corresponding to the one-step period of the user is extracted from the acquired acceleration data.
From the extracted acceleration data, a one-dimensional ideal waveform that serves as a criterion for counting the number of steps of the user is set.
A walking data processing program characterized by executing processing. An acquisition unit that acquires the timing of landing or taking off for each step of the user,
An acquisition unit that acquires acceleration data indicating changes over time,
An extraction unit that extracts acceleration data corresponding to the period of one step of the user from the acquired acceleration data by using the acquired landing or takeoff timing.
From the extracted acceleration data, a setting unit that sets a one-dimensional ideal waveform that serves as a judgment criterion for counting the number of steps of the user, and a setting unit.
A walking data processing device characterized by having.

Images