JP2012053829A - User state estimation device and user state estimation method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a user state estimation device and a user state estimation method for estimating the state of a user who sits down in a chair.SOLUTION: A user state estimation device 10 includes: a first interface 101 for acquiring sound pressure data from a microphone arranged at a position where it is possible to collect a sound due to the operation of a user; a second interface 102 for acquiring acceleration data from a three-axial acceleration sensor installed in a chair; a control unit 103 for executing the estimation processing of the user state; and a determination tree storage unit 105 for storing a determination tree to be used for the estimation processing of the user state. The control unit 103 includes: a featured value extraction section 1032 for extracting a plurality of types of parameters as featured values from the sound pressure data and the acceleration data in a prescribed period; and a user state estimation section 1033 for reading the determination tree from the determination tree storage part 105, and for executing the estimation processing of at least 5 types of the user states based on the extracted featured values.

Description

  The present invention relates to a user state estimation device and a user state estimation method for estimating the state of a user sitting on a chair.

  A technique is known in which an acceleration sensor is installed on a swivel chair and a context of a user seated on the swivel chair is estimated (see, for example, Non-Patent Document 1). This technique is effective in that the user does not have to put on a special sensor, but the user states that can be estimated are limited to “sitting” and “leaving”.

  On the other hand, in order to estimate the operation of a user who operates a personal computer (PC) on a desk, in order to estimate the user's operation in a more detailed manner by enlarging the user state that can be estimated than the technique of Non-Patent Document 1, A technique for mounting an acceleration sensor on an arm or the like is known (for example, see Non-Patent Document 2). In this technique, four states of “PC work”, “reading a document”, “writing a document”, and “free action” can be estimated.

Masafumi Okubo, Ango Fujimura, “Proposal of Concentration Estimation System Using Accelerometer”, [online], [Search August 1, 2010], Internet http://www.wiss.org/WISS2008Proceedings/posters/ paper0038.pdf Makoto Sato, Chie Morita, Miwako Doi, “Recognition of Behavior Using Biological Data and Acceleration Data”, [online], [searched on August 1, 2010], Internet http://www.ipsj.or.jp/01kyotsu /award/taikai_yushu/65award_papers/Y_Sato_Makoto.pdf

  In the conventional method of installing an acceleration sensor on a swivel chair in Non-Patent Document 1, there are only two user-sitting states that can be estimated, and when this is applied to a service, there is only a very limited use method. could not.

  Further, in the conventional method using the wearable acceleration sensor in Non-Patent Document 2, there is a problem that places a burden on the user in that the sensor must always be worn by the user even when seated. Further, the user status related to device use such as “PC work” can be estimated more easily by monitoring the PC operation itself than using data collected from sensors such as keyboard / mouse operations.

  For this reason, the technique which estimates more types of user states, without mounting | wearing a user himself / herself with a sensor is desired.

  The present invention has been made in view of the above-described problems, and it is an object of the present invention to provide a user state estimation device and a user state estimation method for estimating the state of a user sitting on a chair.

  The present invention extracts a plurality of types of parameters as feature quantities from data for a certain period collected from a triaxial acceleration sensor installed on a chair and a microphone installed on a desk, and sits on the chair based on these parameters. The state of the user who performs is estimated.

  In order to reduce the burden on the user, an acceleration sensor installed on a chair and a microphone (for example, a directional microphone) placed on a desk are used as sensors. The acceleration sensor collects vibrations caused by the movement of the user sitting on the chair, and the microphone collects sounds near the user sitting on the chair. A plurality of types of parameters as feature quantities are extracted based on data collected from both the acceleration sensor and the microphone, and each of various user states at the time of seating determined in advance (obtained by learning in advance) with high accuracy is obtained. judge.

  That is, the user state estimation device of the present invention is a user state estimation device that estimates the state of a user sitting on a chair, and the sound pressure from a microphone disposed at a position where sound caused by the user's motion can be collected. A first interface for acquiring data; a second interface for acquiring acceleration data from a three-axis acceleration sensor installed in the chair; a control unit for executing processing for estimating the user state; and estimating the user state A decision tree storage unit for storing a decision tree used for processing, wherein the control unit monitors acceleration data and sound pressure data obtained via the first interface and the second interface, and performs a predetermined period of time. A data monitoring unit for determining the acceleration data and sound pressure data of the sound, and a plurality of types as feature amounts from the sound pressure data and acceleration data of the predetermined period A feature quantity extraction unit that extracts parameters; and a user state estimation unit that reads out a decision tree from the decision tree storage unit and executes at least five types of user state estimation processing based on the extracted feature quantities. It is characterized by that.

  Further, in the user state estimation device of the present invention, the plurality of types of parameters as the feature amount include an average value, a variance value, and a power spectrum of each axis obtained from the acceleration data, and an average value, variance of the three axes combined And a parameter consisting of a value and a power spectrum, and a parameter consisting of an average value of sound pressure obtained from sound pressure data, a variance value and a power spectrum.

  In the user state estimation apparatus according to the present invention, the decision tree includes a decision tree according to a collective estimation method or a decision tree according to a step estimation method.

  Furthermore, the user state estimation method of the present invention is a user state estimation method for estimating the state of a user sitting on a chair, and the sound pressure from a microphone disposed at a position where sound caused by the user's action can be collected. A step of acquiring data, a step of acquiring acceleration data from a three-axis acceleration sensor installed in the chair, and monitoring the acceleration data and sound pressure data to determine acceleration data and sound pressure data for a predetermined period. A step of extracting a plurality of types of parameters as feature values from the sound pressure data and acceleration data of the predetermined period, and at least 5 based on the extracted feature values using a predetermined decision tree Performing a process for estimating types of user states.

  ADVANTAGE OF THE INVENTION According to this invention, the user's state at the time of sitting can be estimated without imposing a burden on a user, and more user states can be estimated compared with the past.

It is a figure which shows the example of application of the user state estimation apparatus of one Example by this invention. 3 is a diagram illustrating an axial direction of a triaxial acceleration sensor 30. FIG. It is a block diagram of the user state estimation apparatus of one Example by this invention. (A), (b) is a figure which shows the example of the sound pressure data and acceleration data of the cut-out period in the user state estimation apparatus of one Example by this invention. It is a figure which shows the operation example which performs the estimation process of at least 5 types of user states from the extracted feature-value using the decision tree according to the collective estimation method in the user state estimation apparatus of one Example by this invention. It is a figure which shows the operation example which performs the estimation process of at least 5 types of user states from the extracted feature-value using the decision tree according to the stage estimation system in the user state estimation apparatus of one Example by this invention. (A), (b) is a figure which shows the standard deviation of the estimated correct answer rate finally obtained by the step estimation method and the collective estimation method, respectively, and an estimated correct answer rate, respectively. It is a figure which shows the example of the decision tree of a batch estimation system (the 1). It is a figure which shows the example of the decision tree of a batch estimation system (the 2). It is a figure which shows the example of the decision tree of a batch estimation system (the 3). It is a figure which shows the example of the decision tree of a batch estimation system (the 4). It is a figure which shows the example of the decision tree of a batch estimation system (the 5). It is a figure which shows the example of the decision tree of a stage estimation system (the 1). It is a figure which shows the example of the decision tree of a stage estimation system (the 2). It is a figure which shows the example of the decision tree of a stage estimation system (the 3). It is a figure which shows the example of the decision tree of a stage estimation system (the 4). It is a figure which shows the example of the decision tree of a stage estimation system (the 5). It is a figure which shows the example of the decision tree of a stage estimation system (the 6). It is a figure which shows the example of the decision tree of a stage estimation system (the 7). It is a figure which shows the example of the decision tree of a stage estimation system (the 8). It is a figure which shows the example of the decision tree of a stage estimation system (the 9). It is a figure which shows the example of the decision tree of a stage estimation system (the 10). It is a figure which shows the example of the decision tree of a stage estimation system (the 11).

  Hereinafter, a user state estimation apparatus according to an embodiment of the present invention will be described. The user state estimation method according to the present invention will become clear from the description of the user state estimation device.

  FIG. 1 is a diagram illustrating an application example of a user state estimation apparatus according to an embodiment of the present invention. On the desktop 50, a display device 40 and a microphone 20 connected to the user state estimation device 10 are installed. The user U is seated on the rotary chair 60. A triaxial acceleration sensor 30 is attached to the backrest portion of the swivel chair 60, and the triaxial acceleration sensor 30 is capable of short-range wireless communication (for example, Bluetooth) with the user state estimation device 10. FIG. 2 shows a state in which a triaxial acceleration sensor (sampling frequency: 10 Hz) is installed on the backrest of the chair and a directional microphone is installed on the desk. The axial direction of the triaxial acceleration sensor 30 is shown in FIG. Show. Note that the microphone 20 does not have to be on the desk as long as it can pick up sound due to the operation of the user U.

  FIG. 3 is a block diagram of a user state estimation apparatus according to an embodiment of the present invention. The user state estimation device 10 of the present embodiment includes an interface (I / F) 101 that acquires sound pressure data from the microphone 20, and an interface (I / F) 102 that acquires acceleration data from the triaxial acceleration sensor 30. A control unit 103 that executes a user state estimation process, an output unit 104 that outputs the user state estimation process to the display device 40, a decision tree storage unit 105 that stores a decision tree used for the user state estimation process, And a storage unit 106. The user state estimation device 10 can be configured as a computer, and stores a program describing processing contents for realizing each function of the control unit 103 in the storage unit 106 of the computer, and performs central processing of the computer. This can be realized by reading and executing this program by a device (CPU).

  The control unit 103 includes a data monitoring unit 1031, a feature amount extraction unit 1032, and a user state estimation unit 1033.

  The data monitoring unit 1031 monitors acceleration data and sound pressure data obtained via the I / Fs 101 and 102, and either one or both of the acceleration data and the sound pressure data has changed more than a certain level. In this case, time-axis data for a period (corresponding to a window function period) until the level at which the change occurs is stabilized is cut out and sent to the feature amount extraction unit 1032. For example, FIG. 4A shows an example of sound pressure data in the cut out period, and FIG. 4B shows an example of acceleration data in the cut out period.

  The feature amount extraction unit 1032 extracts a plurality of types of parameters as feature amounts from the sound pressure data and acceleration data in the cut out period, and sends them to the user state estimation unit 1033. For example, from the acceleration data from the triaxial acceleration sensor 30, the average value, variance value, and power spectrum of each axis, and the average value, variance value, and power spectrum of the triaxial composite axis are extracted as feature amounts. Further, from the sound pressure data from the microphone 20, the average value, variance value, and power spectrum of the sound pressure are extracted as feature amounts. In this embodiment, the sampling frequency is 8000 Hz, the number of bits is 16 bits, the data size of the window function (FFT) is 8192 points, and 4096 points that are effective as a power spectrum (half of the FFT data size), 800 in the first half. Only points are used for estimation.

  The user state estimation unit 1033 reads the decision tree according to the batch estimation method or the decision tree according to the step estimation method from the decision tree storage unit 105 that stores the decision tree according to the batch estimation method or the decision tree according to the step estimation method, Based on the feature amount extracted by the feature amount extraction unit 1032, at least five types of user state estimation processes are executed, and the estimation results are sent to the display device 40 via the output unit 104. In this embodiment, the five states of “break”, “conversation”, “reading”, “writing”, and “food” are the center of gravity position / the voice of the person / the sound of turning the paper / the sound of running the pen / tableware. It is possible to estimate with high accuracy by paying attention to the sound to be made.

  FIG. 5 shows an operation example in which at least five types of user state estimation processes are executed from extracted feature quantities using a decision tree according to the collective estimation method. The batch estimation is an estimation method using a decision tree of merged clusters, and is a method of estimating a context using a pre-constructed decision tree with all extracted feature values as input values. The sound pressure data and acceleration data are acquired by the data monitoring unit 1031 and cut out (step S1), the feature values of the sound pressure data and acceleration data are extracted by the feature value extraction unit 1032 (step S2), and the user state estimation unit In 1033, five states of "rest", "conversation", "reading", "writing", and "food" are estimated using a decision tree according to the collective estimation method (step S3).

  FIG. 6 is a diagram illustrating an operation example in which at least five types of user state estimation processing are executed from extracted feature quantities using a decision tree according to the stage estimation method in the user state estimation apparatus according to an embodiment of the present invention. is there. The stage estimation method is an example using a decision tree along a two-stage cluster. A method for estimating a context step by step using a plurality of decision trees built in advance using feature values extracted from sensors as input values. It is. The sound pressure data and acceleration data are acquired by the data monitoring unit 1031 and cut out (step S11), and the feature values of the sound pressure data and acceleration data are extracted by the feature value extraction unit 1032 (step S12), and the user state estimation unit. 1033, first, using the decision tree (acceleration) according to the stage estimation method, the user states of “with motion” and “without motion” are roughly classified (step S13). Using the decision tree (sound pressure data from the microphone) according to the stage estimation method, five states of “rest”, “conversation”, “reading”, “writing”, and “food” are estimated (step S14).

  FIG. 7A shows the estimated accuracy rates finally obtained for both formulas, and FIG. 7B shows the standard deviation of the estimated accuracy rates finally obtained for both formulas. As can be seen from the results of FIG. 7A, the collective estimation method obtains a high estimated accuracy rate in all remaining contexts except the “conversation” context. Compared to this, the stage estimation method has no operation, and the “conversation” context, which is characterized by speech, has a good result due to the characteristics of the method.

  It can be seen from the standard deviation value shown in FIG. 7B that the value of the collective estimation method is small in all contexts. In other words, this indicates that the influence of individual differences by the user is small, and it can be said that the method can be applied without complicated customization.

  Examples of decision trees for the batch estimation method are shown in FIGS. 8A to 8E, and examples of decision trees for the stage estimation method are shown in FIGS. 9A to 9K. The explanation of symbols in the exemplified decision tree is as follows.

(3-axis acceleration sensor)
XAve: X-axis average value
YAve: Y-axis average value
ZAve: Z-axis average value
ALLAve: Composite axis average value of each axis
XVar: X-axis dispersion value
YVar: Y-axis variance
ZVar: Z-axis dispersion value
ALLVar: Composite axis dispersion value for each axis
XPow: X-axis power spectrum
YPow: Y-axis power spectrum
ZPow: Z-axis power spectrum
ALLPow: Composite axis power spectrum of each axis (microphone)
MAve: Average sound pressure
MVar: Dispersion value of sound pressure
MPow1: First point of the output power spectrum
MPow800: 800th point of output power spectrum
MPowAve: Average value of 800 power spectra in the first half
MPowVar: Dispersion value for 800 power spectra in the first half (estimated context)
condition = 1: Reading
condition = 2: writing
condition = 3: Break
condition = 4: Food and drink
condition = 5: conversation

  Thus, according to the present invention, it is possible to estimate the user's state at the time of sitting without burdening the user, and it is possible to estimate more user states than in the past.

  As an example of the application of this user status, when receiving a call to a seated user, it becomes possible to select an optimal communication means (for example, a telephone or an e-mail) for the user according to the estimated user status and to receive the incoming call. . Since the user's state can be estimated in detail, application to an action history and a life log can be expected.

  According to the present invention, since various types of user states can be automatically recognized with high accuracy, the present invention is useful for environmental facilities that require control according to the user state.

DESCRIPTION OF SYMBOLS 10 User state estimation apparatus 20 Microphone 30 3-axis acceleration sensor 40 Display apparatus 50 Desktop 60 Chair 101,102 Interface (I / F)
DESCRIPTION OF SYMBOLS 103 Control part 104 Output part 105 Decision tree storage part 106 Storage part 1031 Data monitoring part 1032 Feature-value extraction part 1033 User state estimation part

Claims (4)

A user state estimation device for estimating a state of a user sitting on a chair,
A first interface for acquiring sound pressure data from a microphone arranged at a position where sound caused by a user's operation can be collected;
A second interface for acquiring acceleration data from a three-axis acceleration sensor installed in the chair;
A control unit that executes the estimation process of the user state;
A decision tree storage unit for storing a decision tree used for the estimation process of the user state,
The controller is
A data monitoring unit that monitors acceleration data and sound pressure data obtained via the first interface and the second interface, and determines acceleration data and sound pressure data for a predetermined period;
A feature amount extraction unit that extracts a plurality of types of parameters as feature amounts from the sound pressure data and acceleration data of the predetermined period;
A user state estimation unit that reads out a decision tree from the decision tree storage unit and executes at least five types of user state estimation processing based on the extracted feature amount;
A user state estimation device comprising: The plural types of parameters as the feature amount are:
Parameters consisting of the average value, variance value, and power spectrum of each axis obtained from the acceleration data, and the average value, variance value, and power spectrum of the three-axis composite axis,
The user state estimation apparatus according to claim 1, further comprising: a parameter composed of an average value, a variance value, and a power spectrum of sound pressure obtained from the sound pressure data.   The user state estimation apparatus according to claim 2 or 3, wherein the decision tree includes a decision tree according to a batch estimation method or a decision tree according to a step estimation method. A user state estimation method for estimating a state of a user sitting on a chair,
Obtaining sound pressure data from a microphone arranged at a position where sound caused by the user's action can be collected;
Obtaining acceleration data from a three-axis acceleration sensor installed in the chair;
Monitoring the acceleration data and sound pressure data to determine acceleration data and sound pressure data for a predetermined period;
Extracting a plurality of types of parameters as feature values from the sound pressure data and acceleration data of the predetermined period;
Executing at least five types of user state estimation processing based on the extracted feature quantity using a predetermined decision tree;
The user state estimation method characterized by including.

Images