JP2015029696A - Similarity calculation device, similarity calculation method, similarity calculation program, and information processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a similarity calculation device or the like calculating similarity at high speed even when the number of pieces of data increases.SOLUTION: A similarity calculation device 101 has: a characteristic amount calculation part 102 calculating a characteristic amount characterizing data according to a prescribed characteristic amount calculation procedure; a range evaluation part 103 calculating an overlapping degree representing similarity related to the characteristic amount according to a prescribed degree calculation procedure on the basis of a first characteristic amount that is the characteristic amount calculated by the characteristic amount calculation part 102 about data included in teacher data and a second characteristic amount that is the characteristic amount calculated by the characteristic amount calculation part 102 about target data; a selection part 104 selecting specific data from the teacher data according to the overlapping degree; and a similarity calculation part 105 calculating similarity between the specific data and the target data.

Description

  The present invention relates to a similarity calculation device for calculating a similarity between data.

  An information processing apparatus such as a smartphone or a mobile terminal includes an acceleration sensor that measures an acceleration related to a user's operation or the like. A technique for estimating a user's motion based on an acceleration measured by an acceleration sensor is applied to a field of operating a device according to a game, a gesture, or a motion.

  For example, Patent Literature 1 discloses a technique for estimating a user's motion using an acceleration sensor. In the method disclosed in Patent Document 1, the similarity related to the operation is calculated based on the acceleration measured by the two acceleration sensors. First, the user moves the apparatus having the acceleration sensor by holding it in his hand. Next, the acceleration sensor measures acceleration related to the user's motion. Further, the apparatus estimates the user's motion based on data relating to acceleration measured by the acceleration sensor (hereinafter referred to as “acceleration data”).

  The technique disclosed in Patent Literature 1 associates in advance acceleration data measured when a user performs each motion with respect to a plurality of motions to be estimated (hereinafter, motions to be estimated) and the types of the motions. Based on the obtained data (hereinafter referred to as “teacher data”), the type of operation performed by the user is estimated.

  First, the acceleration sensor measures the acceleration related to the user's operation. Next, the technique disclosed in Patent Document 1 calculates a similarity between the acceleration data and the teacher data. As a result, the technique estimates the motion associated with the teacher data having the highest similarity with the acceleration data as the user motion.

  For example, the teacher data includes acceleration data measured when “waving the arm”, “turning the arm”, and the like. In the technique, acceleration data is acquired by an acceleration sensor measuring a user's movement. And this technique compares the acceleration data measured with each acceleration data in teacher data based on the system which patent document 1 discloses. For example, when the acceleration data measured by the acceleration sensor is more similar to the acceleration data measured when “waving the arm” than the acceleration data measured when “turning the arm”, the technique is It is presumed that the motion is a “waving arm” motion.

  For example, the human body motion determination device disclosed in Patent Document 2 is a device that estimates a user's motion based on acceleration. The human body motion determination device compares acceleration data measured by the acceleration sensor with acceleration data previously associated with the type of user's motion. The human body motion determination device estimates the type of motion performed by the user by referring to the user motion associated with the acceleration data having a small difference between the two acceleration data.

JP 2010-187282 A JP 2000-213967 A

  The human body motion discrimination device disclosed in Patent Document 2 acquires acceleration data for each estimation target motion, and creates teacher data by collecting the acquired acceleration data. Therefore, the number of teacher data increases as the types of estimation target motions increase. As a result, when the human body motion determination device disclosed in Patent Document 2 is used, the number of times that the acceleration data measured by the acceleration sensor and the acceleration data in the teacher data are compared increases, so the time for estimating the type of motion is become longer.

  Accordingly, a main object of the present invention is to provide a similarity calculation device capable of calculating a similarity at high speed even when the number of data related to an operation to be estimated increases.

  In order to achieve the above-described object, a similarity calculation apparatus according to the present invention is characterized by having the following configuration.

That is, the similarity calculation device according to the present invention is
A feature amount calculation unit that calculates a feature amount that characterizes the data according to a predetermined feature amount calculation procedure;
Based on the first feature value that is the feature value calculated by the feature value calculation unit for data included in the teacher data and the second feature value that is the feature value calculated by the feature value calculation unit for target data. In accordance with a predetermined frequency calculation procedure, a range evaluation unit that calculates the degree of overlap representing the degree of similarity related to the feature amount;
A selection unit that selects specific data from the teacher data according to the degree of overlap;
A similarity calculation unit that calculates the similarity between the specific data and the target data is provided.

As another aspect of the present invention, the similarity calculation method according to the present invention includes:
Using the information processing apparatus, the first feature amount, which is a feature amount calculated according to a predetermined feature amount calculation procedure for data included in the teacher data, and the feature calculated according to the predetermined feature amount calculation procedure for target data Based on the second feature value that is a quantity, according to a predetermined frequency calculation procedure, calculate the degree of overlap representing the degree of similarity with respect to the feature quantity, and select specific data from the teacher data according to the degree of overlap, The similarity between the specific data and the target data is calculated.

  Furthermore, this object is also realized by such a similarity calculation program and a computer-readable recording medium for recording the program.

  According to the similarity calculation device and the like according to the present invention, the similarity can be calculated at high speed even when the number of data increases.

It is a block diagram which shows the structure which the similarity calculation apparatus which concerns on the 1st Embodiment of this invention has. It is a flowchart which shows the flow of a process in the similarity calculation apparatus which concerns on 1st Embodiment. It is a figure showing an example of the predetermined feature-value calculation procedure. It is a figure showing an example of the acceleration which the acceleration sensor which concerns on 1st Embodiment measures. It is a figure showing an example of the local maximum value which the feature-value calculation part which concerns on 1st Embodiment detects. It is a figure showing an example which divides the field about the 2nd value at the 1st interval. It is a block diagram which shows the structure which the similarity calculation apparatus which concerns on the 2nd Embodiment of this invention has. It is a flowchart which shows the flow of the process in the feature-value calculation part which concerns on 2nd Embodiment. It is a figure showing an example which divides | segments time at a 2nd space | interval. It is a block diagram which shows the structure which the similarity calculation apparatus which concerns on the 3rd Embodiment of this invention has. It is a flowchart which shows the flow of a process in the feature-value calculation part which concerns on 3rd Embodiment. It is a block diagram which shows the structure which the similarity calculation apparatus which concerns on the 4th Embodiment of this invention has. It is a flowchart which shows the flow of a process in the similarity calculation apparatus which concerns on 4th Embodiment. It is a figure which represents conceptually an example of the process in which the change part which concerns on 4th Embodiment changes a 2nd space | interval recursively. It is explanatory drawing showing the structural example of teacher data in a table form It is a block diagram which shows the structure which the information processing apparatus which concerns on the 5th Embodiment of this invention has. It is a flowchart which shows the flow of a process in the information processing apparatus which concerns on 5th Embodiment. It is a block diagram which shows roughly the hardware constitutions of the calculation processing apparatus which can implement | achieve the similarity calculation apparatus which concerns on each embodiment of this invention.

  Next, embodiments for carrying out the present invention will be described in detail with reference to the drawings.

<First Embodiment>
The configuration of the similarity calculation apparatus 101 according to the first embodiment of the present invention and the processing performed by the similarity calculation apparatus 101 will be described in detail with reference to FIGS. 1 and 2. FIG. 1 is a block diagram showing the configuration of the similarity calculation apparatus 101 according to the first embodiment of the present invention. FIG. 2 is a flowchart showing the flow of processing in the similarity calculation apparatus 101 according to the first embodiment.

  The similarity calculation apparatus 101 according to the first embodiment includes a feature amount calculation unit 102, a range evaluation unit 103, a selection unit 104, and a similarity calculation unit 105.

  For example, while using an example in which an information processing apparatus such as a smartphone includes an acceleration sensor that measures an acceleration representing a user's action and the similarity calculation apparatus 101 according to the present embodiment, the flow of processing in the similarity calculation apparatus 101 Will be described.

  First, for example, a user performs a plurality of operations that are estimation targets while holding an information processing apparatus such as a smartphone in advance. At the same time, the acceleration sensor measures acceleration (for example, FIG. 4) related to the user's motion. FIG. 4 is a diagram illustrating an example of acceleration measured by the acceleration sensor.

  In FIG. 4, the horizontal axis indicates that the time (in the present embodiment, it is not limited to the time and is hereinafter also referred to as “first value”) in the direction of the arrow. The vertical axis indicates that the acceleration (in the present embodiment, it is not limited to acceleration, and is hereinafter also referred to as “second value”) increases in the direction of the arrow.

  The teacher data is, for example, measured acceleration (hereinafter referred to as “acceleration data”. In addition, in the present embodiment, the teacher data is also referred to as “target data” since it is not limited to acceleration), and the type of the operation. Associate. The acceleration data in the teacher data is associated with the feature amount calculated according to a predetermined feature amount calculation procedure. Further, the teacher data may further include information relating to the age and sex of the person who has operated in association with the acceleration data. The similarity calculation apparatus 101 according to the present embodiment calculates the similarity based on the above-described feature amount and the like.

  The predetermined feature amount calculation procedure is, for example, the procedure shown in FIG. FIG. 3 is a diagram illustrating an example of a predetermined feature amount calculation procedure. That is, the feature amount calculation unit 102 calculates a feature amount related to data according to a predetermined feature amount calculation procedure.

  First, the feature amount calculation unit 102 searches for a local maximum value related to the second value (acceleration in the example shown in FIG. 4) in the acceleration data (step S301). For example, in the case of the acceleration data shown in FIG. 4, the feature quantity calculation unit 102 sets the maximum values of acceleration shown in FIG. 5 (circles in the figure) (for example, the times at which the acceleration becomes maximum, respectively, t1, t2 , T3, t4, and t5). FIG. 5 is a diagram illustrating an example of the maximum value detected by the feature amount calculation unit 102. For example, the feature amount calculation unit 102 calculates the inclination of the acceleration data, and searches for the maximum value according to the calculated inclination.

  Next, the feature amount calculation unit 102 divides the region composed of the second values (acceleration data in the example shown in FIG. 4) at the first interval (FIG. 6), and the local maximum value in each divided section. Calculate the frequency. FIG. 6 is a diagram illustrating an example of dividing the region relating to the second value at the first interval. In FIG. 6, for example, the region having the second value is divided into four sections having a first section to a fourth section by the first interval.

  The first interval may be determined in advance, or may be calculated based on the maximum value and the minimum value of the second value, for example. Further, the first interval does not necessarily have to have the same width in each section.

  Next, the feature quantity calculation unit 102 calculates the frequency at which the maximum value appears in each section (step S302). In the example shown in FIG. 6, the first interval, the second interval, and the third interval are each one (maximum value at time t3), one (maximum value at time t1), and three (time t2, t4). And the maximum value at t5).

  Next, the feature amount calculation unit 102 calculates the first value (in the example shown in FIG. 6, time) that gives a local maximum value with a low frequency (however, 1 or more) in the example of FIG. 6 as the feature amount. (Step S303). For example, in the example illustrated in FIG. 6, since the first section and the second section are smaller than the number of maximum values in the third section, the feature amount calculation unit 102 calculates time t1 and time t3 as the feature amount.

  The predetermined feature amount calculation procedure includes, for example, a procedure (or expressed as “method”) shown in the following embodiments. The predetermined feature amount calculation procedure may be any value, code, or the like calculated based on the data, and is not limited to the above-described procedure.

  In the above-described example, the feature amount calculation unit 102 calculates the time when the maximum value appears as the feature amount, but the time has a certain range (that is, between the start time and the end time in the section). It may be the center time in the section. That is, the feature amount is not limited to the value in the above-described example.

  First, the feature quantity calculation unit 102 receives, for example, acceleration data measured by an acceleration sensor, and represents a feature quantity (hereinafter referred to as “second feature quantity”) representing the characteristics of the received acceleration data according to a predetermined feature quantity calculation procedure. (Step S201).

  Next, the range evaluation unit 103 follows the predetermined frequency calculation procedure based on the feature amount (hereinafter referred to as “first feature amount”) related to each data included in the teacher data and the second feature amount. The degree of overlap, which is the degree of similarity between the two feature quantities, is calculated (step S202).

  As the predetermined frequency calculation procedure, for example, as shown in Equation 1, there is a method of calculating the degree of overlap according to the degree of discreteness between the first feature value and the second feature value. For example, when the first feature value is tt1, tt2,... Ttn in order from the earliest time and the second feature value is tu1, tu2,... Tun, the range evaluation unit 103 calculates the overlap degree P according to Equation 1. .

P = 1 ÷ (sum of | ttk−tuk | with respect to k) (Expression 1),
(Where | x | represents the absolute value of x).

  The predetermined frequency calculation procedure is not limited to the above-described example as long as the similarity between the first feature value and the second feature value is calculated.

  Next, the selection unit 104 selects specific data from the teacher data in accordance with the degree of overlap calculated by the range evaluation unit 103 according to a predetermined frequency calculation procedure (step S203). For example, the selection unit 104 selects one or more data represented by the first feature value having a large degree of overlap with the second feature value as the specific data from the teacher data.

  Next, the similarity calculation unit 105 calculates the similarity between the specific data selected by the selection unit 104 and the acceleration data (step S204). For example, the similarity calculation unit 105 calculates the similarity between acceleration data and specific data according to the method disclosed in Patent Document 1.

  In the present embodiment, specific data having a large degree of overlap with the second feature value that characterizes the acceleration data is selected from the teacher data, and the similarity with the acceleration data is calculated for the selected specific data. That is, according to this embodiment, the number of teacher data to be compared with acceleration data can be reduced. Therefore, according to the similarity calculation apparatus according to the first embodiment, the similarity can be calculated at high speed even when the number of data increases.

  On the other hand, for example, in the human body motion determination device disclosed in Patent Document 2, the number of teacher data increases as the types of motions to be estimated increase. As a result, the number of data to be compared with the acceleration data increases in the human body movement determination device as the number of teacher data increases. Therefore, when the data to be compared with the acceleration data increases, it is difficult for the human body motion determination device to calculate the similarity at high speed.

  In the above description, the smartphone has an acceleration sensor that measures acceleration, but the present embodiment is not limited to the above example. For example, in the present embodiment, when determining a medical condition from a waveform in an electrocardiogram, when analyzing a meaning from a waveform representing a voice, or when estimating whether or not the device fails from a temperature change in the device, etc. It can be applied to various situations. That is, this embodiment is not limited to the above-described example.

  In the above-described example, the information processing apparatus has an acceleration sensor that measures acceleration and a function that estimates an operation based on the measured acceleration. However, one information processing apparatus does not necessarily have the above two functions. There is no need. In other words, the acceleration may be measured by one information processing apparatus, and the operation related to the measured acceleration may be estimated by another information processing apparatus. That is, this embodiment is not limited to the above-described example.

<Second Embodiment>
Next, a second embodiment of the present invention based on the first embodiment described above will be described.

  In the following description, the characteristic part according to the present embodiment will be mainly described, and the same components as those in the first embodiment described above will be denoted by the same reference numerals, and redundant description will be omitted. To do.

  The configuration of the similarity calculation device 701 according to the second embodiment and the processing performed by the similarity calculation device 701 will be described with reference to FIGS. FIG. 7 is a block diagram showing a configuration of a similarity calculation apparatus 701 according to the second embodiment of the present invention. FIG. 8 is a flowchart showing the flow of processing in the feature amount calculation unit 702 according to the second embodiment.

  The similarity calculation device 701 includes a feature amount calculation unit 702, a range evaluation unit 103, a selection unit 104, and a similarity calculation unit 105.

  The feature amount calculation unit 702 calculates a feature amount related to received data according to a predetermined feature amount calculation procedure (steps S801 to S804).

  For example, the feature amount calculation unit 702 divides the region composed of the first values into one or more sections according to the second interval (step S801). For example, as in the example illustrated in FIG. 9, when the first value represents time, the feature amount calculation unit 702 divides the time at the second interval (the section created by the division is expressed as “time window "). FIG. 9 is a diagram illustrating an example of dividing the time at the second interval. In the example shown in FIG. 9, the time is divided into first to fourth time windows.

  Next, the feature amount calculation unit 702 calculates a second value representing the divided section as a representative value in accordance with a predetermined representative value calculation procedure (step S802). For example, as a predetermined representative value calculation procedure, there is a method of calculating a value such as a maximum value, a minimum value, an average value, or a variance value of the second value in each time window. Since various methods for calculating the representative value already exist, description on the method is omitted here.

  Next, the feature amount calculation unit 702 divides the region made up of the second value (acceleration in the example shown in FIG. 9) into one or more sections at the third interval, and calculates in each divided section. The frequency relating to the representative value is calculated (step S803).

  For example, the third interval may be determined in advance, or may be calculated based on the maximum value and the minimum value of the second value, for example. The third interval does not necessarily have to have the same width in each section.

  Next, the feature amount calculation unit 702 calculates a first value (in the example illustrated in FIG. 9) that gives a representative value with a low frequency (however, 1 or more) as a feature amount (step S804).

  In the first embodiment, the frequency related to the maximum value is calculated, but in the present embodiment, the frequency related to the representative value is calculated.

  Since the second embodiment includes the same configuration as that of the first embodiment, the same effect as that of the first embodiment can be obtained. That is, according to the similarity calculation apparatus according to the second embodiment, the similarity can be calculated at high speed even when the number of data increases.

  By using an average value or a variance value, the types of representative values can be increased. As a result, since various features can be captured, according to the present embodiment, the similarity can be calculated with higher accuracy than in the first embodiment.

<Third Embodiment>
Next, a third embodiment of the present invention based on the first embodiment described above will be described.

  In the following description, the characteristic part according to the present embodiment will be mainly described, and the same components as those in the first embodiment described above will be denoted by the same reference numerals, and redundant description will be omitted. To do.

  The configuration of the similarity calculation device 901 according to the third embodiment and the processing performed by the similarity calculation device 901 will be described with reference to FIGS. 10 and 11. FIG. 10 is a block diagram showing a configuration of a similarity calculation apparatus 901 according to the third embodiment of the present invention. FIG. 11 is a flowchart showing a flow of processing in the feature amount calculation unit 902 according to the third embodiment.

  The similarity calculation device 901 includes a feature amount calculation unit 902, a range evaluation unit 103, a selection unit 104, and a similarity calculation unit 105.

  The feature amount calculation unit 902 calculates a feature amount related to data according to a predetermined feature amount calculation procedure (steps S1001 to S1005).

  For example, the feature amount calculation unit 902 divides the region including the first value into one or more sections according to the second interval (step S1001). For example, as in the example illustrated in FIG. 9, when the first value represents time, the feature amount calculation unit 902 divides the time at the second interval (the section created by the division is referred to as “time window ").

  Next, the feature amount calculation unit 902 performs Fourier transform on each function (that is, a function that associates the first value with the second value) in the created section (for example, “time window”) (step S1002). ). The function is converted into a function for associating the frequency and the amplitude by Fourier transform.

  Next, the feature amount calculation unit 902 calculates a frequency associated with a large amplitude in the created section (step S1003). In the case of the example illustrated in FIG. 9, each time window is associated with a specific frequency by the feature amount calculation unit 902 calculating the frequency.

  Next, the feature amount calculation unit 902 divides the region composed of the second value (in this example, the frequency) into one or more sections at the fourth interval, and calculates the frequency calculated in each divided section. The frequency regarding is calculated (step S1004).

  For example, the fourth interval may be determined in advance, or may be calculated based on the maximum value and the minimum value of the second value, for example. Further, the fourth interval does not necessarily have to have the same width in each section.

  Next, the feature amount calculation unit 902 calculates a first value (in the example shown in FIG. 9, time) giving a low frequency (however, 1 or more) frequency as a feature amount (step S1005).

  Since the third embodiment includes the same configuration as that of the first embodiment, the third embodiment can enjoy the same effects as those of the first embodiment. That is, according to the similarity calculation apparatus according to the third embodiment, the similarity can be calculated at high speed even when the number of data increases.

  In the present embodiment, by calculating a frequency having a large amplitude using Fourier transform, for example, fine fluctuations in the user's operation can be removed. As a result, according to the present embodiment, an error or the like when measuring the acceleration can be reduced. For this reason, the similarity can be calculated with higher accuracy than in the first embodiment described above.

<Fourth Embodiment>
Next, a fourth embodiment of the present invention based on the second embodiment will be described.

  In the following description, the characteristic part according to the present embodiment will be mainly described, and the same reference numerals will be given to the same configurations as those of the second embodiment described above, thereby omitting redundant description. To do.

  The configuration of the similarity calculation device 1101 according to the fourth embodiment and the processing performed by the similarity calculation device 1101 will be described with reference to FIGS. FIG. 12 is a block diagram showing a configuration of a similarity calculation device 1101 according to the fourth embodiment of the present invention. FIG. 13 is a flowchart showing the flow of processing in the similarity calculation device 1101 according to the fourth embodiment.

  The similarity calculation device 1101 includes a feature amount calculation unit 102, a range evaluation unit 103, a selection unit 1104, a similarity calculation unit 1105, and a change unit 1106.

  First, as described in the first embodiment, the selection unit 1104 selects specific data from the teacher data according to the degree of overlap calculated by the range evaluation unit 103 according to a predetermined frequency calculation procedure (step S1201). .

  Next, the changing unit 1106 determines whether or not the specific number of data is smaller than a predetermined threshold (step S1202). When determining that the number of specific data is larger than the predetermined threshold (NO in step S1202), the changing unit 1106 narrows the second interval (step S1203).

  The procedure for setting the predetermined threshold includes, for example, a method of setting a positive value such as 3 or 10 in advance and a method of setting a value according to the processing performance of the information processing apparatus. The procedure for setting the predetermined threshold is not limited to the example described above.

  Next, the feature amount calculation unit 102 calculates a second feature amount based on the second interval narrowed by the changing unit 1106 (step S1204).

  Next, the range evaluation unit 103 calculates a degree of overlap, which is a degree of similarity related to the feature amount, according to a predetermined frequency calculation procedure based on the first feature amount and the second feature amount (step S1205).

  When the changing unit 1106 determines that the number of specific data is smaller than the predetermined threshold (YES in step S1202), steps S1203 to S1205 are not performed.

  In the above description, the similarity calculation device 1101 performs the processing from step S1201 to step S1205. However, the processing may be recursively repeated. For example, as illustrated in FIG. 14, when the second interval is half the size of the area having the first value, the changing unit 1106 repeatedly performs the process of further halving the area in each section. May be. FIG. 14 is a diagram conceptually illustrating an example of processing in which the changing unit 1106 recursively changes the second interval.

  Note that the method of dividing the region does not need to divide the region into two halves, and may be divided into N (N is an integer of 3 or more).

  Since the fourth embodiment includes the same configuration as the second embodiment, the fourth embodiment can enjoy the same effects as those of the second embodiment. That is, according to the similarity calculation apparatus 1101 according to the fourth embodiment, the similarity can be calculated at high speed even when the number of data increases.

  The changing section 1106 narrows the second interval, so that the created section becomes small. Therefore, the feature amount calculation unit 102 calculates a representative value in a smaller section. The range evaluation unit 103 calculates the degree of overlap in a smaller section based on the representative value calculated by the feature amount calculation unit 102. As the section is reduced, the number of sections having a large degree of overlap decreases. Therefore, according to the present embodiment, the number of specific data decreases. That is, according to the present embodiment, the similarity can be calculated at a higher speed.

  An example in which an information processing apparatus such as a smartphone includes the similarity calculation apparatus 1101 according to the present embodiment and an acceleration sensor that measures a user's operation will be described. The teacher data includes four types of operations related to “projecting an arm”, “turning an arm”, “waving an arm”, and “pulling an arm” as operations for which the similarity calculation device 1101 calculates a similarity, for example. Suppose that it contains motion data.

  First, the user holds the smartphone and presses an “operation start” button displayed on the display of the smartphone. The acceleration sensor measures acceleration data in response to the user pressing a button.

  On the other hand, the smartphone displays a message such as “sensor data acquisition start” and an “operation end” button on the display. The user confirms the display of a message such as “start sensor data acquisition”, and then performs an operation that is the basis of the target data. In response to the completion of the operation, the user presses an “end operation” button displayed on the display. The acceleration sensor ends the measurement of acceleration data in response to the user pressing the “end operation” button.

  In response to the user pressing the “end operation” button, the similarity calculation device 1101 calculates the similarity based on the acceleration data. The smartphone estimates the type of motion related to the acceleration data according to the similarity calculated by the similarity calculation device 1101.

  However, for example, acceleration occurs when the user presses a button. The acceleration data measured by the acceleration sensor is data having an error because it is different from the acceleration data measured in the operation intended by the user. The smartphone may have a function of separating acceleration data that is generated unintentionally, for example, when the user presses a button. That is, the smartphone may extract acceleration data measured when the user performs an action intended from the acceleration data measured by the acceleration sensor.

  In the following description, a case where the smartphone estimates the type of the user's operation based on the acceleration data measured by the acceleration sensor when the user performs the “protruding arm” operation is taken as an example.

  The feature amount calculation unit 102 calculates the second feature amount according to a predetermined feature amount calculation procedure based on the acceleration data measured by the acceleration sensor. For example, the feature amount calculation unit 102 calculates the value of a peak (for example, an inflection point in an upward convex function) as the second feature amount based on the acceleration data. Assume that the acceleration sensor measures acceleration data in FIG. As shown in FIG. 5, there are five peaks in the acceleration data in the acceleration data. In the following description, these five points are represented as a, b, c, d, and e in order of time.

  Next, the feature quantity calculation unit 102 specifies, for example, a peak representing a characteristic change in the obtained five peaks. Specifically, as shown in FIG. 6, the feature amount calculation unit 102 first divides an area made up of acceleration data into one or more sections based on the first interval. For example, the feature amount calculation unit 102 divides the region between the maximum value and the minimum value of the acceleration data into four sections as the first interval. For the sake of explanation, the divided sections are expressed as a first section, a second section, a third section, and a fourth section in order from the smallest acceleration data.

  Next, the feature amount calculation unit 102 calculates the number of times the peak appears in each of the first to fourth intervals. In the case of the example shown in FIG. 6, the first section, the second section, and the third section are one (peak c), one (peak a), and three (peak b, peak d, peak e), respectively. Includes peaks.

  Next, the feature amount calculation unit 102 calculates, as the feature amount, the first value (in the example shown in FIG. 6, time) that gives a low frequency (however, 1 or more) peak in the example of FIG. For example, the feature amount calculation unit 102 calculates the time t1 when the peak a occurs and the time t3 when the peak c occurs as the feature amounts.

  For example, in the above-described example, the length of time for which the acceleration sensor measures acceleration data is 1000 (millisecond, milli second, hereinafter referred to as [msec]), the measurement start time is 0 [msec], and t1 is 160. It is assumed that [msec] and t3 are 310 [msec].

  The teacher data is assumed to be data shown in FIG. FIG. 15 is an explanatory diagram illustrating a configuration example of teacher data in a table form. In the example illustrated in FIG. 15, the teacher data associates the feature amount, the length of time to be measured (hereinafter referred to as “measurement time”), and the type of operation. For example, the operation of “extending an arm” is associated with the feature quantity “150, 300” and the measurement time “100”.

  The teacher data further includes, for example, time length information of each teacher data, environment information for acquiring the teacher data (for example, information such as the age and gender of the person who has operated), performance information regarding the acceleration sensor (for example, sampling frequency) Or resolution). In this case, the similarity calculation device 1101 calculates the similarity based also on the information described above.

  The range evaluation unit 103 calculates the degree of overlap between the feature amount calculated by the feature amount calculation unit 102 based on the acceleration data and the feature amount in the teacher data according to a predetermined frequency calculation procedure.

  As an example of the predetermined frequency calculation procedure, a method of dividing the time from when the acceleration data is measured until the measurement is completed will be described.

  The range evaluation unit 103 divides the measurement time of acceleration data (that is, 1000 [msec]) into two to divide it into a first section and a second section. Assume that the first interval is 0 to 500 [msec], and the second interval is 501 to 1000 [msec]. In this case, there are two peaks (peaks at time 160 [msec] and 310 [msec]) in the first section.

  Similarly, the range evaluation unit 103 divides the measurement time for each data in the teacher data. For example, the feature amount of the “extrus arm” data in the teacher data is 150 [msec] and 300 [msec]. Therefore, there are two peaks in the first section. Further, the “turn arm” data has one peak in the second section, and the “pull arm” data has two peaks in the first section. It should be noted that the data of “waving the arm” has no peak in the measurement time.

  The range evaluation unit 103 determines that the “extend arm” data and the “pull arm” data have a two-point peak in the first interval, similarly to the acceleration data. As a result, the range evaluation unit 103 calculates that the acceleration data, the “extend arm” data, and the “pull arm” data have a high degree of overlap.

  The selection unit 1104 selects “extend arm” data and “pull arm” data as specific data according to the degree of overlap calculated by the range evaluation unit 103.

  Next, the changing unit 1106 may determine whether the specific number of data is less than a predetermined threshold.

  For example, when the predetermined threshold is 3, the changing unit 1106 does not change the first interval because it determines that the specific number of data is smaller than the predetermined threshold. In this case, the selection unit 1104 selects “extend arm” data and “pull arm” data as specific data according to the degree of overlap calculated by the range evaluation unit 103.

  When the predetermined threshold is 1, the changing unit 1106 changes the first interval in order to determine that the specific number of data is larger than the predetermined threshold. For example, the changing unit 1106 changes the first interval by dividing each of the first section and the second section into two.

  Thereafter, the first section is divided into a 1-1 section (that is, 0 to 250) and a 1-2 section (that is, 251 to 500), and the second section is divided into a 2-1 section (that is, 501). To 750) and 2-2 section (that is, 751 to 1000).

  Since peaks in the acceleration data exist at times 160 [msec] and 310 [msec], the range evaluation unit 103 determines that there is one peak in the 1-1 section and one peak in the 1-2 section. To do. Further, the range evaluation unit 103 determines that there is one peak in the 1-1 section and one peak in the 1-2 section for the “extend arm” data.

  Similarly, the range evaluation unit 103 determines that there are two peaks in the first-second section for the “pull arm” data. That is, the range evaluation unit 103 calculates a higher degree of overlap in the “extend arm” data than in the “pull out arm” data, since the peak appearing sections match in the “extrude arm” data and the acceleration data. To do. In this case, the selection unit 1104 selects the “projecting arm” data as specific data according to the degree of overlap calculated by the range evaluation unit 103.

  The similarity calculation unit 1105 calculates the similarity between the specific data and the acceleration data. The smartphone determines the type of operation related to the acceleration data according to the similarity calculated by the similarity calculation unit 1105. For example, when the similarity calculation unit 1105 determines that the similarity between the acceleration data and the “extend arm” data is high, the smartphone displays on the display that the user's operation is an “extrude arm” operation. .

  That is, according to this embodiment, the number of teacher data to be compared with acceleration data can be reduced. Therefore, according to the similarity calculation apparatus according to the present embodiment, the similarity can be calculated at high speed even when the number of data increases.

<Fifth Embodiment>
Next, a fifth embodiment of the present invention based on the first embodiment described above will be described.

  In the following description, the characteristic part according to the present embodiment will be mainly described, and the same components as those in the first embodiment described above will be denoted by the same reference numerals, and redundant description will be omitted. To do.

  The configuration of the information processing apparatus 1508 according to the fifth embodiment and the processing performed by the information processing apparatus 1508 will be described with reference to FIGS. FIG. 16 is a block diagram showing a configuration of an information processing apparatus 1508 according to the fifth embodiment of the present invention. FIG. 17 is a flowchart showing the flow of processing in the information processing apparatus 1508 according to the fifth embodiment.

  The information processing device 1508 includes a similarity calculation device 1501, a determination unit 1507, and a management unit 1506. The similarity calculation device 1501 includes a feature amount calculation unit 1502, a range evaluation unit 1503, a selection unit 104, and a similarity calculation unit 105.

  The determination unit 1507 calculates a value associated with specific data with a high similarity (in the example shown in FIG. 14 and the like, the type of action) according to the similarity calculated by the similarity calculation unit 105. For example, when the acceleration data is most similar to the “extend arm” data, the determination unit 1507 estimates that the user's action is “extend the arm”.

  The management unit 1506 determines whether or not to update the teacher data depending on whether or not the value estimated by the determination unit 1507 matches the received value (step S1601). If the management unit 1506 determines that the value estimated by the determination unit 1507 matches the received value (YES in step S1601), the management unit 1506 does not update the teacher data. When determining that the value estimated by the determination unit 1507 does not match the received value (NO in step S1601), the management unit 1506 updates the value based on the received teacher data (step S1602).

  For example, it is assumed that the management unit 1506 receives a received value (that is, “extends an arm” in this example). When the determination unit 1507 estimates that the acceleration data is “extends the arm” data, the value received (that is, “extends the arm” in this example) matches the value estimated by the determination unit 1507. For this reason, the management unit 1506 does not update the teacher data.

  On the other hand, for example, it is assumed that the management unit 1506 receives a received value (that is, “pull an arm” in this example). In this case, the value estimated by the determination unit 1507 (that is, “extends the arm” in this example) and the value received (that is, “pulls the arm” in this example) are different. Accordingly, the management unit 1506 updates the teacher data based on the received value (that is, “pull the arm” in this example). That is, the management unit 1506 associates the received value with the feature amount calculated by the feature amount calculation unit 1502 based on the acceleration data, and adds it to the teacher data. The management unit 1506 may delete the feature amount or the like associated with the value estimated by the determination unit 1507 (in this example, “protrud arm”) from the teacher data.

  Processing in the information processing apparatus 1508 according to the present embodiment will be described using the example shown in the fourth embodiment.

  The information processing device 1508 (for example, a smartphone) has a “correct” button for sending a signal indicating that the value estimated by the determination unit 1507 is correct to the management unit 1506 on a display (not shown), and the determination unit 1507 estimates A “wrong” button for sending a signal indicating that the value to be performed is an error to the management unit 1506 is displayed. The user presses the “correct answer” button when the value estimated by the determination unit 1507 is correct. On the other hand, when the value estimated by the determination unit 1507 is incorrect, the user presses the “wrong” button.

  In response to a process in which the user presses the “correct answer” button or the “wrong” button, the management unit 1506 starts the process.

  When the user presses the “wrong” button, the information processing apparatus 1508 displays an input field for inputting a character string on a display (not shown). The user inputs the type of operation in the input field. Next, the management unit 1506 associates the type of motion input by the user with the feature amount calculated based on the acceleration data and adds it to the teacher data.

  For example, it is assumed that the user inputs “turn the arm” in the input field, and the feature amounts calculated by the feature amount calculation unit 1502 regarding the acceleration data are 160 [msec], 300 [msec], and 800 [msec]. In this example, the management unit 1506 adds “turn the arm” to the teacher data in association with the feature amounts of 160 [msec], 300 [msec], and 800 [msec]. The management unit 1506 may delete the teacher data associated with the erroneous value.

  Since the similarity calculation device according to the fifth embodiment includes the same configuration as that of the first embodiment, the same effect as that of the first embodiment can be obtained. That is, according to the similarity calculation apparatus according to the fifth embodiment, the similarity can be calculated at high speed even when the number of data increases.

  Furthermore, when the value estimated by the determination unit 1507 is not correct, the management unit 1506 updates the teacher data associated with the value to a correct value. Therefore, the teacher data correctly associates the acceleration data corresponding to the user's operation measured by the acceleration sensor and the type of the operation. As a result, the range evaluation unit 1503 calculates the degree of overlap based on the teacher data updated by the management unit 1506. That is, since the degree of overlap calculated by the range evaluation unit 1503 becomes more accurate, the information processing apparatus according to the present embodiment can further correctly determine the user's action.

(Hardware configuration example)
A configuration example of hardware resources that realizes the similarity calculation device according to each embodiment of the present invention described above using one calculation processing device (information processing device, computer) will be described. However, the similarity calculation device may be realized using at least two calculation processing devices physically or functionally. Further, the similarity calculation device may be realized as a dedicated device.

  FIG. 18 is a diagram schematically illustrating a hardware configuration of a calculation processing apparatus capable of realizing the similarity calculation apparatus according to the first to fifth embodiments. The calculation processing device 20 includes a central processing unit (Central Processing Unit, hereinafter referred to as “CPU”) 21, a memory 22, a disk 23, a nonvolatile recording medium 24, an input device 25, and an output device 26.

  The non-volatile recording medium 24 can be read by a computer, for example, a compact disc (Compact Disc), a digital versatile disc, a Blu-ray Disc (registered trademark), a universal serial bus memory (USB memory). ), A solid state drive (Solid State Drive), etc., and the program can be retained and carried even without power supply. The nonvolatile recording medium 24 is not limited to the above-described medium. Further, the program may be carried via a communication network instead of the nonvolatile recording medium 24.

  That is, the CPU 21 copies a software program (computer program: hereinafter simply referred to as “program”) stored in the disk 23 to the memory 22 when executing it, and executes arithmetic processing. The CPU 21 reads data necessary for program execution from the memory 22. When the display is necessary, the CPU 21 displays the output result on the output device 26. When inputting a program from the outside, the CPU 21 reads the program from the input device 25. The CPU 21 calculates a similarity calculation program (FIGS. 2 and 3) in the memory 22 corresponding to the function (process) represented by each unit shown in FIG. 1, FIG. 7, FIG. 10, FIG. 12, or FIG. 8, FIG. 11, FIG. 13 or FIG. 17) are interpreted and executed. The CPU 21 sequentially performs the processes described in the above-described embodiments of the present invention.

  That is, in such a case, it can be understood that the present invention can also be realized by such a similarity calculation program. Furthermore, it can be understood that the present invention can also be realized by a computer-readable recording medium in which the similarity calculation program is recorded.

DESCRIPTION OF SYMBOLS 101 Similarity calculation apparatus 102 Feature quantity calculation part 103 Range evaluation part 104 Selection part 105 Similarity calculation part 701 Similarity calculation apparatus 702 Feature quantity calculation part 901 Similarity degree calculation apparatus 902 Feature quantity calculation part 1101 Similarity degree calculation apparatus 1104 Selection part DESCRIPTION OF SYMBOLS 1105 Similarity calculation part 1106 Change part 1501 Similarity calculation apparatus 1502 Feature-value calculation part 1503 Range evaluation part 1506 Management part 1507 Determination part 1508 Information processing apparatus 20 Computation processing apparatus 21 CPU
22 Memory 23 Disk 24 Non-volatile recording medium 25 Input device 26 Output device

Claims (10)

A feature amount calculation unit that calculates a feature amount that characterizes the data according to a predetermined feature amount calculation procedure;
Based on the first feature value that is the feature value calculated by the feature value calculation unit for data included in the teacher data and the second feature value that is the feature value calculated by the feature value calculation unit for target data. In accordance with a predetermined frequency calculation procedure, a range evaluation unit that calculates the degree of overlap representing the degree of similarity related to the feature amount;
A selection unit that selects specific data from the teacher data according to the degree of overlap;
A similarity calculation device comprising: a similarity calculation unit that calculates the similarity between the specific data and the target data. When the second value is determined according to the first value, the feature amount calculation unit divides the region including the second value into a plurality of first sections according to a first interval, and the second value Calculating the local maximum value of the value, measuring the frequency at which the calculated local maximum value appears for the first section, and calculating the first value giving the local maximum value with the low frequency as the feature amount The similarity calculation device according to claim 1, wherein the feature amount is calculated by a feature amount calculation procedure. When the second value is determined according to the first value, the feature amount calculation unit divides the region including the first value into a plurality of second sections according to a second interval, and calculates a predetermined representative value According to the procedure, a representative value representative of the second interval is calculated, and the region including the calculated representative value is divided into a plurality of third intervals according to a third interval, and the representative value appears in the third interval. The feature amount is calculated by the predetermined feature amount calculation procedure for measuring the frequency and calculating the first value in the second section that gives the representative value with the low measured frequency as the feature amount. Item 3. The similarity calculation device according to item 1. The feature amount calculation unit divides a region including the first value into a plurality of second sections according to a second interval in a function in which a second value is determined according to the first value, Fourier transform of the function, the frequency corresponding to a large amplitude value obtained after the transformation is divided based on the fourth interval, the frequency in the divided fourth interval is measured, and the second interval giving the low frequency The similarity calculation device according to claim 1, wherein the feature amount is calculated by the predetermined feature amount calculation procedure for calculating the first value in the as the feature amount. The range evaluation unit calculates the degree of overlap according to the predetermined frequency calculation procedure for calculating the degree of overlap according to a degree of discreteness between the first feature amount and the second feature amount. The similarity calculation apparatus according to claim 2. According to the number of data in the specific data, further comprising a change unit that calculates the second interval,
The similarity calculation apparatus according to any one of claims 3 to 5, wherein the predetermined feature amount calculation procedure is a procedure of calculating the feature amount based on the second interval calculated by the changing unit. Each of the teacher data is associated with an action type,
The similarity calculation device according to claim 1,
An information processing apparatus comprising: a determination unit that calculates a type of the operation associated with data having the largest similarity. The information processing apparatus according to claim 7, further comprising: a management unit that associates the target data with a correct operation type and adds to the teacher data when the operation type calculated by the determination unit is incorrect.   Using the information processing apparatus, the first feature amount, which is a feature amount calculated according to a predetermined feature amount calculation procedure for data included in the teacher data, and the feature calculated according to the predetermined feature amount calculation procedure for target data Based on the second feature value that is a quantity, according to a predetermined frequency calculation procedure, calculate the degree of overlap representing the degree of similarity with respect to the feature quantity, and select specific data from the teacher data according to the degree of overlap, A similarity calculation method for calculating a similarity between the specific data and the target data. A feature amount calculation function for calculating a feature amount that characterizes data according to a predetermined feature amount calculation procedure;
Based on the first feature value that is the feature value calculated by the feature value calculation function for data included in the teacher data and the second feature value that is the feature value calculated by the feature value calculation function for target data. A range evaluation function for calculating the degree of overlap representing the degree of similarity related to the feature amount according to a predetermined frequency calculation procedure;
A selection function for selecting specific data from the teacher data according to the degree of overlap;
A similarity calculation program for causing a computer to realize a similarity calculation function for calculating the similarity between the specific data and the target data.

Images