JP2011045524A - Device for estimating exercise intensity and/or calorie consumption of carrier - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device for estimating exercise intensity with high accuracy, with no need to fix in a predetermined direction. <P>SOLUTION: The estimating device includes: a means for setting a plurality of transform periods so that their starting points differ from each other in a predetermined number and subjecting the accelerations included in the respective transform periods to Fourier transform to obtain a plurality of spectra, and substituting each of the plurality of obtained spectra with the representative spectrum with the least error among a plurality of predetermined representative spectra; a means for obtaining appearance probabilities in the plurality of representative spectra obtained by the substitution for each of a plurality of characteristic spectra selected from the plurality of representative spectra in advance, and generating data for estimation including the appearance probabilities of the obtained respective characteristic spectra; a means for determining states corresponding to the data for estimation based on the study data including the same data as the data for estimation measured at least one time in the respective states; and a means for deciding the exercise intensity based on the determined states. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a technique for estimating exercise intensity and calorie consumption of a person who possesses the device by using an acceleration sensor mounted on the device.

  Various proposals have been made on portable devices for estimating exercise intensity and calorie consumption in daily life (see, for example, Patent Documents 1, 2, and 3).

  According to Patent Document 1, the exercise intensity is estimated based on the standard deviation value of the output value of the acceleration sensor included in the estimation device over a certain period. According to Patent Document 2, the estimation device includes a triaxial acceleration sensor, and calculates an average value of the posture component and the motion component of the combined acceleration from the output value in each axis direction of the triaxial acceleration sensor, and calculates these values. From this value, it is determined whether the user is currently running, walking, climbing stairs, descending stairs, standing, sitting, or standing, and energy consumption is calculated based on the determined status. ing. Further, according to Patent Document 3, the estimation apparatus includes a triaxial acceleration sensor, calculates a representative value in each axial direction from an output value of each triaxial acceleration sensor in a certain time, and represents a representative horizontal component. The exercise intensity is estimated based on the value and the representative value of the vertical component.

JP 2006-204446 A JP 2007-160076 A JP 2009-028312 A

  For example, the exercise intensity differs between a state where the body is stationary and a state where the body is running, but the method described in Patent Document 1 does not determine the state of the user, and the measured acceleration is constant. The exercise intensity is estimated by using only the standard deviation value in the period, and there is a problem that the estimation accuracy is low. In addition, the estimation devices described in Patent Documents 2 and 3 are inconvenient because it is necessary to fix the three-axis acceleration sensor to the user so that each axis is in a predetermined direction. In particular, when considering that the estimation device is implemented as a device having other functions, for example, a device integrated with a mobile communication terminal, other functions cannot be used due to fixing of the estimation device to the user. Or the problem that it becomes difficult to use arises. That is, in this case, the configurations described in Patent Documents 2 and 3 cannot be employed.

  Further, for example, in the state of riding a bicycle, a train, an automobile, etc., the acceleration sensor detects accelerations other than the user's own motion. However, the above-described conventional technology is a state where the user uses some vehicle. Therefore, there is a problem that the estimation accuracy in the state where the vehicle is used deteriorates.

  Therefore, an object of the present invention is to provide an apparatus for estimating exercise intensity and / or calorie consumption, which does not need to be fixed and held in a predetermined direction and has higher estimation accuracy than the prior art.

According to the exercise intensity estimation device of the present invention,
Among the accelerations measured by the acceleration measuring means and the acceleration measuring means, the predetermined number of conversion periods are set within the estimation period of the predetermined length while the start positions of the conversion periods of the predetermined length are shifted from each other. The Fourier transform means for obtaining a plurality of spectra by Fourier transforming the accelerations included in each conversion period, and each of the plurality of spectra obtained by the Fourier transform means within a plurality of predetermined representative spectra. For each of a plurality of representative spectrums selected in advance from the plurality of representative spectrums, a quantization means for substituting the representative spectrum with the smallest representative spectrum is obtained, An estimation data generator that generates estimation data including the appearance probability of each feature spectrum And classifying means for determining a state corresponding to the estimation data based on the learning data, having learning data including the same data as the estimation data, measured at least once in each state And exercise intensity determining means for determining exercise intensity based on the above.

According to another embodiment of the estimation device of the present invention,
It is also preferable that a means for obtaining a variance value of acceleration is provided for each of the predetermined number of conversion periods, and the estimation data includes an average value and a variance value of the obtained prescribed number of variance values. .

Further, according to another embodiment of the estimation device of the present invention,
The estimation data preferably includes information indicating a frequency having the maximum amplitude in a plurality of representative spectra obtained by the replacement by the quantization means.

Furthermore, according to another embodiment of the estimation device of the present invention,
It is also preferable that a means for acquiring position information is provided, and the classification means uses the position information for determining the state.

Furthermore, according to another embodiment of the estimation device of the present invention,
The position information is information indicating a base station that is communicating by wireless communication, and the classification unit preferably uses the change frequency of the base station to determine the state.

Furthermore, according to another embodiment of the estimation device of the present invention,
It is also preferable that the exercise intensity determining means uses an average value of the dispersion values for determining the exercise intensity.

Furthermore, according to another embodiment of the estimation device of the present invention,
Means for collecting operation information about operations on the device, and the exercise intensity determining means, when an operation on the device is performed, out of the acceleration measured by the acceleration measuring means before and after the operation is performed; It is also preferable to apply a filter to acceleration for a certain period, calculate an average value of the variance values from the filtered acceleration, and use it for determination of exercise intensity.

Furthermore, according to another embodiment of the estimation device of the present invention,
The state determined by the classification unit includes a state where the vehicle is riding on a train or a car, and the exercise intensity determination unit is configured to detect the acceleration measured by the acceleration measuring unit when the state is a state where the vehicle is riding on a train or a car. It is also preferable to apply a filter and calculate an average value of the variance values from the filtered acceleration and use it for the determination of exercise intensity.

Furthermore, according to another embodiment of the estimation device of the present invention,
It is also preferable that the exercise intensity determining means uses information indicating a representative spectrum that appears most frequently among a plurality of representative spectra obtained by the quantization means replacing the exercise intensity.

Furthermore, according to another embodiment of the estimation device of the present invention,
It is also preferable that the classifying unit includes a plurality of classifiers connected in multiple stages, and the classifier at each stage determines a further classified state from the state determined by the classifier at the previous stage.

According to the calorie consumption estimation device of the present invention,
The calorie consumption is calculated based on the exercise intensity determined by the exercise intensity estimating device.

  In the present invention, the representative spectrum sequence is obtained from the acceleration within the estimation period, and the state is determined based on the appearance probability of the feature spectrum in each state measured in advance and the appearance probability of the feature spectrum in the obtained representative spectrum sequence. Therefore, it is possible to prevent the influence of a peculiar spectrum that may be observed at a certain moment, and to accurately determine the state based on the learning data measured in advance. In the present invention, since information about the direction of acceleration is not used for estimation, it is not necessary to fix the device to the user.

1 is a simplified configuration diagram of an apparatus for estimating exercise intensity and / or calorie consumption according to the present invention. FIG. It is a block diagram of a data converter. It is a figure explaining a data collection method. It is a figure which shows the measured spectrum series. It is a figure which shows a representative spectrum table. It is a figure which shows prior data. It is a figure explaining the determination of a characteristic spectrum. It is a figure which shows the characteristic spectrum table. It is a figure which shows learning data. It is a figure explaining determination of exercise intensity. It is a figure which shows other embodiment of the estimation apparatus by this invention.

  EMBODIMENT OF THE INVENTION The form for implementing this invention is demonstrated in detail below using drawing.

  The estimation device according to the present invention, for example, indicates the state of a user carrying the estimation device as “stop state”, “walking state”, “running state”, “state driving a bicycle”, “train” Or, the exercise intensity and calorie consumption are estimated. FIG. 1 is a simplified configuration diagram of an estimation apparatus according to the present invention. The estimation apparatus includes an acceleration sensor 1, a composite acceleration calculation unit 2, a data conversion unit 3, a classifier 4, and an exercise intensity determination unit 5. And a calorie consumption calculation unit 6.

The acceleration sensor 1 is a triaxial acceleration sensor in the present embodiment, periodically measures accelerations in three directions orthogonal to each other, and outputs the measured values to the combined acceleration calculation unit 2. The combined acceleration calculation unit 2 calculates the combined acceleration at each time, that is, the actual magnitude of the acceleration, from the three directions of acceleration simultaneously collected by the acceleration sensor 1 by the following formula.
Composite acceleration = (x ² + y ² + z ² ) ^0.5 (1)
Here, x, y, and z are accelerations in the respective directions.

  The data conversion unit 3 generates estimation data for estimating the state from the combined acceleration at each time calculated by the combined acceleration calculation unit 2 and outputs the estimation data to the classifier 4. FIG. 2 is a configuration diagram of the data conversion unit 3. The data conversion unit 3 includes a Fourier transform unit 31, a variance calculation unit 32, a quantization unit 33, a representative spectrum table 34, and an estimation data generation unit. 35 and a characteristic spectrum table 36.

  As shown in the upper part of FIG. 3, the combined acceleration at each acceleration measurement time is input to the data conversion unit 3, and the Fourier transform unit 31 includes a plurality of FFT periods. The resultant acceleration included in the L FFT periods of −L is Fourier transformed to calculate the spectrum of the resultant acceleration. Note that the start position of each FFT period is set by being shifted by the shift period Ts as shown in FIG. Further, a period including all FFT periods is referred to as an estimation period as shown in FIG. As shown in FIG. 4, the Fourier transform unit 31 includes a spectrum 70-1 obtained from the combined acceleration included in the FFT period 60-1, a spectrum 70-2 calculated from the combined acceleration included in the FFT period 60-2, and the last. In addition, a spectrum series including the spectrum 70 -L obtained from the combined acceleration included in the FFT period 60 -L is output to the quantization unit 33.

The quantization unit 33 selects and selects a spectrum closest to a predetermined reference from a plurality of representative spectra described in the representative spectrum table 34 for each of the input spectra 70-1 to 70-L. The spectrums 70-1 to 70-L are replaced with the representative spectrum. FIG. 5 is a diagram showing the representative spectrum table 34. The generation of the representative spectrum table 34 will be described later. According to FIG. 5, the representative spectrum table 34 has a total of N representative spectra to which identifiers S _{1 to} S _N are assigned. For example, the representative spectrum to which S ₁ is assigned as an identifier is It is shown that the DC component (frequency number 0) is 3, and the first, second, and third frequency components are 6, 1, and 9, respectively.

The quantization unit 33 replaces each of the input spectra 70-1 to 70-L with the representative spectrum that is closest, in other words, with the least error, and outputs a spectrum sequence after replacement. Here, the error is evaluated by, for example, a value obtained by summing the absolute value of the amplitude difference of the corresponding frequency component or the square of the amplitude difference over all frequencies. For example, the closest representative spectrum to the spectrum 70-1 is _{S 34,} if the closest representative spectrum to the spectrum 70-2 is _{S 184,} the closest representative spectrum to a spectrum 70-L is _{S 10,} the quantization The unit 33 outputs the replacement representative spectrum identifier series to the estimation data generation unit 35 as in (S ₃₄ , S ₁₈₄ ,..., S ₁₀ ). In addition, this replacement operation is to aggregate a plurality of spectrum states into one representative spectrum, which is one kind of quantization. As a result, the measured spectrum series can be expressed by the representative spectrum identifier series.

  In addition, the variance calculation unit 32 calculates the variance value of the composite acceleration included in each FFT period, and outputs it to the estimation data generation unit 35. That is, a total of L variance values are output to the estimation data generation unit 35.

  The estimation data generation unit 35 generates estimation data to be output to the classifier 4 using the feature spectrum table 36 based on the spectrum series from the quantization unit 33 and the variance value from the variance calculation unit 32. . Here, the characteristic spectrum is a representative spectrum that is important for distinguishing each state in the representative spectrum, and the appearance probability is predetermined in each state from a large number of measurement data measured in each state in advance. The representative spectrum determined to be higher than the value is used as the feature spectrum. Details of the characteristic spectrum determination will be described later. FIG. 8 is a diagram showing the feature spectrum table 36, and a total of seven representative spectra are selected as feature spectra by prior measurement.

As shown in FIG. 8, when seven feature spectra are set, in this embodiment, the estimation data generation unit 35 generates estimation data including nine data. The first data of the estimated data is in the L representative spectrum input, a probability of occurrence of the first aspect spectrum S _21. Second data of the estimated data is in the L representative spectrum input, a probability of occurrence of the second feature spectrum S _31. Hereinafter, similarly, data up to the seventh of the estimation data is calculated.

Further, the eighth data of the estimation data is an average value (hereinafter referred to as V average value) of L variance values from the variance calculation unit 32, and the last data is from the variance calculation unit 32. Of the L variance values (hereinafter referred to as V variance values). That is, for example, in the L representative spectrum from the quantizing unit _33, the probability of occurrence of _{S 21} is _0.3, the probability of occurrence of _{S 31} is _0.2, the appearance probability of 0.1 of _{S 10,} the _{S 66} appearance probability is the appearance probability 0 probability of occurrence is 0, S ₉ occurrence probability is 0, S ₃₃ of 0, S _72, the mean value and variance values of L variance from the variance calculator 32, In the case of 6 and 4, the estimation data generation 35 outputs (0.3, 0.2, 0.1, 0, 0, 0, 0, 6, 4). The reason why the dispersion value of the composite acceleration is obtained instead of the average of the composite acceleration within each FFT period and the average value and the dispersion value over the estimation period of the dispersion value is used is to remove the gravitational acceleration component. In the present invention, the dispersion value of the composite acceleration is used as a value representing the strength of the composite acceleration. That is, the V average value is a value indicating the average of the strength of the combined acceleration within the estimation period.

  The classifier 4 is, for example, a naive Bayes classifier or a naive Bayes tree classifier, and is based on one or more learning data for each state created in advance and set in the classifier 4. A state corresponding to the estimation data input from 3 is determined. FIG. 9 is a diagram showing learning data set in the classifier 4. In FIG. 9, for simplicity, there are three states, “A state”, “B state”, and “C state”. In FIG. 9, the first learning data is measured in the A state, and the appearance probability, V average value, and V variance value of each of the seven feature spectra shown in FIG. 8 are shown. The generation of learning data will be described later.

  The classifier 4 compares the input estimation data with each learning data, selects the learning data having the highest likelihood of the estimation data, and sets the state corresponding to the selected learning data to the exercise intensity determination unit 5. Output to. The exercise intensity determination unit 5 receives L representative spectra and V average values used for generating the estimation data from the data conversion unit 3, and the combined acceleration calculation unit 2 generates the estimation data. Used synthetic acceleration is input.

  The exercise intensity determination unit 5 determines a Mets value based on each state and the V average value. Further, when the state is walking and running, the Met's value varies depending on the number of steps per unit time. Therefore, the exercise intensity determination unit 5 determines the number of steps per unit time based on the resultant acceleration to determine the Met value. To use. Various known methods can be used for determining the number of steps from the combined acceleration. Further, for example, even in the same state, the appearance of different spectra means that there is a difference in state, and naturally the Mets value also differs. For this reason, it is preferable that the representative spectrum that appears most frequently among the L representative spectrums that are the sources of the estimation data is the main spectrum, and the main spectrum is also used to determine the Mets value. In this case, even if the state and the V average value are the same, if the main spectrum is different, different Mets values are output. FIG. 10 shows an example of a table used by the exercise intensity determination unit 5 to determine the Mets value. In FIG. 10, a0 to a2 and b0 to b2 are predetermined coefficients. It should be noted that the relationship between each value and / or formula used to determine the Met's value and the Met's value is obtained by a number of measurements performed in advance.

  Furthermore, when the state is “a state of riding a train or a car”, the exercise intensity determination unit 5 does not use the V average value received from the data conversion unit 3 but inputs from the combined acceleration calculation unit 2 The combined acceleration of the estimated period is filtered, thereby removing an acceleration component caused by vibrations from the vehicle included in the combined acceleration, and obtaining a V average value from the combined acceleration after filtering, It is preferable to use this for conversion to a Mets value.

  Furthermore, when the estimation device according to the present invention is mounted on a device having other functions, such as a mobile communication terminal, an acceleration component due to an operation on the other function, for example, opening / closing of the mobile communication terminal, button pressing, etc. This affects the determination of Mets values. For this reason, a function for acquiring user operation information is provided, and when the user performs some operation, the exercise intensity determination unit 5 filters the combined acceleration for a certain period of time before and after the operation to determine the acceleration due to the operation. It is preferable to remove the component, obtain the V average value from the combined acceleration after filtering, and use this for conversion to a Mets value.

  Finally, the calorie consumption calculating unit 6 obtains calorie consumption by multiplying the Met's value determined by the exercise intensity determining unit 5, the weight of the user previously input by the user, and the duration of the Met value.

  Next, creation of the representative spectrum table 34 will be described. The representative spectrum table 34 corresponds to a quantization step, and the number N of representative spectra is an appropriate number based on the number of spectrum frequencies (K in FIG. 5) and the amplitude values that can be taken for each frequency component. Select. It should be noted that the total sum of errors between representative spectra is preferably selected to be as large as possible so that the representative spectra are not similar. In creating learning data described below, a large number of measurements are performed in each state. A representative spectrum may be determined by a genetic algorithm from a large number of spectra in each state acquired at this time.

  Next, creation of learning data to be set in the feature spectrum table 36 and the classifier 4 will be described. To create the learning data, first, the measurer carries a device having the same size, weight, and shape as the estimation device according to the present invention and including a three-axis acceleration sensor, and in a predetermined state. The acceleration in each direction is collected in the same manner as in the actual estimation, and the resultant acceleration is obtained from Equation (1) from the collected acceleration in each direction. Then, the L spectrum series is obtained from the combined acceleration in the same manner as the actual estimation, and this is replaced with the L representative spectrum series. Further, as in the actual estimation, the V average value and the V variance value are calculated, and the L representative spectrum series, the V average value and the V variance value are recorded as one of the prior data for the state. To do. The acquisition of the prior data is performed a plurality of times for each state, and a plurality of prior data is acquired for each state. At this time, it is preferable that a plurality of measurers carry out the measurement several times for each state, instead of measuring by the same measurer. In addition, the method of possessing the apparatus is also measured in various forms, and the measurement environment, for example, the road state in the case of walking or running, and the vehicle type in the case of riding the vehicle are also measured in various states.

FIG. 6 is a diagram showing prior data. In FIG. 6, for simplicity, there are three states, “A state”, “B state”, and “C state”. According to FIG. 6, the first learning data is measured in the A state, and the L spectral sequences are representative spectrums S ₂₁ , S ₁₀ , S ₈₃ ,..., S _91. It is shown that the V average value and the V variance value are M ₁ and D ₁ , respectively.

Subsequently, a characteristic spectrum is determined from the acquired prior data. FIG. 7 is a diagram for explaining the determination of the characteristic spectrum. First, the appearance probability of each spectrum is calculated for each state of the prior data in FIG. Specifically, for example, if there are 150 pieces of prior data in the A state, since one piece of prior data includes L spectrums, the entire A state includes a total of 150 L representative spectra. For example, if the number of representative spectra S ₂₆ included in the 150 L spectrums is 15, the appearance probability of S ₂₆ is 15 / (150 L). Then, as shown in FIG. 7, the representative spectrums having the appearance probabilities equal to or higher than a predetermined threshold are determined as the feature spectrums in order of increasing appearance probability for each state. For example, when the threshold value is 0.1 in FIG. 7, as shown in FIG. 8, seven representative spectra S ₂₁ , S ₃₁ , S ₁₀ , S ₆₆ , S ₇₂ , S ₃₃ , and S ₉ are selected as feature spectra. Is done. Note that a representative spectrum having almost the same appearance probability in each state is not useful for determination of each state, so even if it has an appearance probability equal to or higher than a threshold, it can be excluded from the feature spectrum.

  Also, the characteristic spectrum can be selected based on the information amount gain of each representative spectrum defined by the following equation.

ここで、ｃは分類すべき各状態、Ｃは全状態、Ｐｒ（Ｓ）はＳの出現確率、Ｐｒ（Ｓ｜ｃ）は状態ｃのときにおけるＳの出現確率、Ｐｒ（ｃ｜Ｓ）はＳが出現したときに状態ｃである確率を示している。具体的には、Ｐｒ（Ｓ｜Ｃ）＞０．０１を満たし、かつ、ＩｎｆｏＧａｉｎ（Ｓ）の上位所定数の代表スペクトラムを特徴スペクトラムとする。

Here, c is each state to be classified, C is all states, Pr (S) is the appearance probability of S, Pr (S | c) is the appearance probability of S in state c, and Pr (c | S) is The probability of being in state c when S appears is shown. Specifically, Pr (S | C)> 0.01 is satisfied, and the upper-level predetermined number of representative spectrums of InfoGain (S) are set as the characteristic spectrum.

Finally, the appearance probability of the characteristic spectrum in the representative spectrum included in each prior data is obtained, and learning data shown in FIG. 9 is created. That is, in the second prior data in FIG. 6, the appearance probabilities of the feature spectra S ₂₁ , S ₃₁ , S ₁₀ , S ₆₆ , S ₇₂ , S ₃₃ , S ₉ are 0.2, 0.19, 0. The second learning data shown in FIG. 9 is generated from the second prior data.

  As described above, in the present invention, in addition to the acceleration dispersion value, the frequency of appearance of each spectrum within the estimation period is used for state determination. Therefore, it is possible to prevent the influence of a peculiar spectrum that may be observed at a certain moment, and to accurately determine the state based on the learning data measured in advance. In the present invention, since information about the direction of each axis of the triaxial acceleration sensor is not used for estimation, it is not necessary to fix the device to the user.

  In the above embodiment, the value based on the appearance probability of the characteristic spectrum and the variance of the acceleration is used as the estimation data, but the frequency having the maximum amplitude value among the L representative spectra output from the quantization unit 33. May be added to the estimation data. This is because the frequency component having the strongest acceleration varies depending on the state, and the estimation accuracy can be further increased. Of course, at this time, the frequency having the maximum amplitude value is also added to the learning data.

  Further, when the estimation apparatus of the present invention is implemented in a mobile communication terminal, information specifying a base station periodically acquired by the mobile communication terminal is supplied to the classifier 4, and the classifier 4 It may be a form used for the determination. Specifically, the frequency of changing the connected base station is converted to a coefficient that increases as the frequency increases, and the likelihood of the learning data “in a train or a car” increases with this coefficient. Adjust as follows. The “stopped state” and “the state of riding in a train or car” may be similar to each other as the appearance probability of the characteristic spectrum, and the “stopped” can be more accurately considered by considering the frequency of base station changes. A “state” can be distinguished from a “state in a train or a car”. Note that if the mobile communication terminal has a function of acquiring position information using GPS, the speed obtained from the acquired position information may be used instead of the frequency of base station change.

  Further, the classifier 4 can have a multi-stage configuration to perform more detailed classification. Specifically, as shown in FIG. 11, four classifiers 41 to 44 are added between the classifier 4 and the exercise intensity determination unit 5. In this embodiment, the classifier 41 is a classifier that determines the detailed state of the walking state, the classifier 42 is a classifier that determines the detailed state of the running state, and the classifier 43 drives the bicycle. The classifier 44 is a classifier that determines the detailed state of a state in which the vehicle is on a train or a car. The classifier 4 outputs the estimation data to the classifier 41 when it is determined from the estimation data as the walking state, and outputs the estimation data to the classifier 42 when it is determined as the running state, and drives the bicycle. When it is determined that the vehicle is in a state of being in a train, the estimation data is output to the classifier 43. If it is, the result of the stop state is output to the exercise intensity determination unit 5.

The classifier 41 classifies the walking state as one of “normal”, “uphill”, “downhill”, “upstairs”, and “downstairs”, and the classifier 42 sets the running state to “normal”. ”,“ Uphill ”,“ downhill ”,“ upstairs ”,“ downstairs ”, and the classifier 43 indicates that the state of driving a bicycle is“ normal ”,“ uphill ” , “Slope descent”, and the classifier 44 determines whether the train or car is in the “sitting position” or “standing position”.
And the detailed state determined by the exercise intensity determination unit 5 is output. Since the exercise intensity determination unit 5 determines the Met's value based on the detailed state, for example, even when the walking state is the same as the number of steps, the normal state, that is, the state walking on a flat road, In the state, different Mets values are calculated.

  As in the classifier 4, learning data for each detailed state measured in advance is set in advance in each of the classifiers 41 to 44. A classifier is configured in a multi-stage connection configuration, and a state that is greatly different first is determined, and then the detailed state is estimated based on the learning data for the state in which each state is further finely classified, thereby accurately estimating the detailed state. It becomes possible.

DESCRIPTION OF SYMBOLS 1 Acceleration sensor 2 Synthetic acceleration calculation part 3 Data conversion part 4, 41, 42, 43, 44 Classifier 5 Exercise intensity determination part 6 Calorie consumption calculation part 31 Fourier transform part 32 Variance calculation part 33 Quantization part 34 Representative spectrum table 35 Estimation Data Generation Unit 36 Feature Spectrum Table 60-1 to 60-L FFT Period 70-1 to 70-L Spectrum

Claims (11)

Acceleration measuring means for measuring acceleration;
Within a predetermined length of the estimation period, a predetermined number of conversion periods of a predetermined length are set with the start positions being shifted from each other, and the acceleration included in each conversion period among the accelerations measured by the acceleration measuring means within the estimation period is calculated. Fourier transform means for obtaining a plurality of spectra by performing Fourier transform respectively,
Quantization means for replacing each of the plurality of spectra obtained by the Fourier transform means with a representative spectrum with the least error among a plurality of predetermined representative spectra,
For each of a plurality of feature spectra selected in advance from the plurality of representative spectra, the appearance probability in the plurality of representative spectra obtained by the replacement by the quantizing means is obtained, and the estimation data including the appearance probability of each obtained feature spectrum is obtained. A data generation means for estimation to be generated;
Classification means for determining the state corresponding to the estimation data based on the learning data, having learning data including the same data as the estimation data, measured at least once in each state;
Exercise intensity determining means for determining exercise intensity based on the determined state;
A device equipped with. In each of the conversion periods set by the predetermined number, a means for obtaining a dispersion value of acceleration is provided.
The estimation data includes an average value and a variance value of the determined predetermined number of variance values,
The apparatus of claim 1. The estimation data includes information indicating a frequency having the maximum amplitude in a plurality of representative spectra obtained by the replacement by the quantization means.
The apparatus according to claim 1 or 2. A means for acquiring location information;
The classifying means uses the position information to determine the state.
The apparatus according to any one of claims 1 to 3. The location information is information indicating a base station communicating by wireless communication,
The classifying means uses the change frequency of the base station to determine the state.
The apparatus according to claim 4. The exercise intensity determining means uses an average value of the variance values for determining the exercise intensity.
The device according to any one of claims 2 to 5. Comprising means for collecting operation information on operations on the device;
The exercise intensity determining means applies a filter to the acceleration for a certain period before and after the operation is performed among the accelerations measured by the acceleration measuring means when an operation is performed on the device, and from the acceleration after the filter , Calculating an average value of the variance values and using it to determine exercise intensity;
The apparatus according to claim 6. The state determined by the classification means includes a state of riding a train or a car,
The exercise intensity determining means applies a filter to the acceleration measured by the acceleration measuring means when the state is in a train or a car, and calculates the average value of the variance values from the filtered acceleration. And used to determine exercise intensity,
Apparatus according to claim 6 or 7. The exercise intensity determination means uses information indicating the representative spectrum that appears most frequently among the plurality of representative spectra obtained by the quantization means for determining the exercise intensity.
Apparatus according to any one of claims 1 to 8. The classification means is a multi-stage connection of multiple classifiers,
The classifier of each stage determines the state further classified from the state determined by the classifier of the previous stage.
Apparatus according to any one of claims 1 to 9.   The apparatus which calculates calorie consumption based on the exercise intensity which the apparatus of any one of Claim 1 to 10 determined.

Images