JP6475589B2 - Identification method, identification apparatus, and identification program for human behavior - Google Patents

Description

  The present invention relates to a human behavior identification method, an identification apparatus, and an identification program. More specifically, the present invention relates to a technique suitable for identifying the type of human behavior in a facility such as a residence and, for example, for watching an elderly person, particularly an elderly person living alone or a disabled person.

  As a conventional technique for confirming the safety of elderly people living alone at home, a plurality of human sensors installed at different locations in the house and emitting detection signals when people move, and detection signals from the human sensors A determination in an abnormal action determination system comprising: a determination means for executing an abnormality determination mode for detecting the start of an action event and determining an abnormal action when the elapsed time from the start of the action event exceeds a threshold value The means further executes a threshold determination mode. In this threshold determination mode, the required time from the start to the end of the action event is measured a number of times, and a frequency distribution with respect to the time axis is obtained from this measurement data, and this frequency distribution data The frequency distribution when the number of measurements is counted from is calculated and the required time for the relative cumulative frequency along the time axis to reach the predetermined value is calculated based on the estimated frequency distribution. And, it is abnormal behavior determining system characterized to determine the computed required time as the threshold value (Patent Document 1).

JP 2001-216585 A

  However, in the behavior determination system of Patent Document 1, although the elapsed time from the start of the behavior can be grasped, the type of behavior actually performed by the person cannot be specified. For this reason, it is hard to say that it is sufficient as a mechanism for watching the elderly and the like because it is impossible to grasp the change in the life pattern of the elderly and the like specifically and in detail after specifying the type of action.

  In the behavior determination system of Patent Document 1, in order to grasp the behavior of elderly people in a residence, it is necessary to cover all locations in the residence without omission within the detection range of each human sensor. Therefore, in many cases, a large number of human sensors are required, which leads to an increase in the cost of constructing a safety confirmation mechanism.

  Accordingly, the present invention provides a human behavior identification method, identification device, and identification program capable of identifying with high accuracy the types of human behavior in a facility such as a residence without necessarily requiring a large number of sensors. The purpose is to provide.

  The present inventor intends to reduce the burden on the resident as much as possible, and while examining a new method for identifying the type of human movement / behavior using a stationary vibration sensor installed in a residence, for example, The following findings were obtained.

  Specifically, the present inventor performs the action of walking a person on the first floor of a wooden house in 55 seconds as a vibration velocity waveform at a sampling frequency of 100 Hz (that is, the measurement time interval is 10 milliseconds). The measurement experiment was conducted twice for the same person.

  As a result of each measurement, a time history waveform (5500 sampling points) shown in FIGS. 3A and 3B was obtained as a result of each measurement. The results shown in FIG. 3 are waveforms obtained by applying a 1 Hz high-pass filter in consideration of the performance of a stationary vibration sensor (specifically, a servo-type speedometer) that measures the speed.

  From the results of (A) and (B) of FIG. 3, it is confirmed that the magnitude of the maximum amplitude and the timing at which the maximum amplitude occurs are greatly different even in the same movement of the same person in the same place. It was.

  In other words, when trying to identify the type of human action / behavior based on the vibration waveform, the same behavior is determined by the time zone to be monitored, the physical condition of the person to be identified, the building structure characteristics, and the performance of the measuring device itself. Even if it was taken, it was thought that some kind of fluctuation occurred in the vibration waveform itself. For this reason, it was considered necessary to identify the type of human movement / behavior while taking into account the degree of fluctuation.

  In view of this, the present inventor considers absorbing fluctuations in human movement / behavior, and calculates the time history waveform (FIGS. 3A and 3B) obtained by the two measurements for each measurement. The absolute velocity spectrum value (that is, spectrum intensity) for each frequency component was calculated by conversion. As a result, the results shown in FIG. 4 (A) and FIG. 4 (B) were obtained. In FIGS. 4 to 6, (A) in each drawing is based on FIG. 3 (A), and (B) in each drawing is based on FIG. 3 (B).

  From the results of (A) and (B) of FIG. 4, if phase information is removed from the waveform data of the time history and converted into a spectrum component, the elapsed time when the maximum amplitude appears if the operation / action is periodic. Regardless of this, it has been found that the frequency components of the maximum amplitude value almost coincide with each other, and similar frequency characteristics appear such that the same frequency component becomes high.

  Further, paying attention to the amplitude value of the time history waveform (FIGS. 3A and 3B) obtained by the two measurements, a frequency distribution of the amplitude value was created for each measurement time. As a result, the results shown in FIG. 5 (A) and FIG. 5 (B) were obtained.

  Furthermore, paying attention to the spectrum value of the spectrum intensity distribution (FIGS. 4A and 4B) obtained by Fourier transform, a frequency distribution of the spectrum value was created for each measurement. As a result, the results shown in FIG. 6 (A) and FIG. 6 (B) were obtained.

  From the results of (A) and (B) in FIG. 5 and (A) and (B) in FIG. 6, the frequency distribution forms in (A) and (B) in each figure are very similar. It has been found that the similarity can also be determined from the viewpoint of the frequency of the amplitude value of each waveform component within the measurement period obtained by removing time information and frequency characteristic information from the waveform data of the time history.

  It has also been found that the frequency distribution shown in FIGS. 5 and 6 can be constructed according to the level of fluctuation to be distinguished by adjusting the width of the horizontal axis.

  For example, based on the above-mentioned knowledge, the present inventors have invented the present invention.

  Specifically, according to the human behavior identification method of the present invention, a frequency distribution of measurement values of reference data acquired as a time history waveform is created by measurement related to operation information for each behavior mode, and the reference data is Fourier transformed. In addition, a frequency distribution of the spectrum values for each frequency component after the Fourier transform is created, and a frequency distribution of the measurement values of the observation data acquired as a time history waveform by the measurement of motion information in the action identification target time zone is created. In addition, the observed data is Fourier transformed and a frequency distribution of spectrum values for each frequency component after the Fourier transformation is created, and thereby, the class attribute of the measured value for each behavior mode related to the reference data and the frequency value Combination data group, combination data group of frequency component and spectrum value, and attribute and degree of spectrum value class Data group, combination data group of measurement value attribute and frequency value of observation data, combination data group of frequency component and spectrum value, and class attribute and frequency of spectrum value A combination data group with values is prepared, and a first determination index value calculated by Equation 1 and Equation 2 for each of the three types of combination data groups relating to each of the reference data and the observation data, Equation 3 and At least two determination index values of the second determination index value calculated by Expression 4 and the third determination index value calculated by Expression 5 are used, and integrated determination is performed for each action mode by Expression 6 to Expression 8. Value is calculated, and the type of human action in the action identification target time zone is identified based on the integrated judgment value for each action mode To have.

  Further, the human behavior identification device of the present invention includes means for creating a frequency distribution of measured values of reference data acquired as a time history waveform by measurement related to operation information for each behavior mode, and means for Fourier transforming the reference data And means for creating a frequency distribution of spectral values for each frequency component after Fourier transform of the reference data, and frequency distribution of measured values of observation data acquired as a time history waveform by measurement related to motion information in the action identification target time zone , A means for Fourier transforming the observation data, and a means for creating a frequency distribution of the spectrum values for each frequency component after the Fourier transformation of the observation data. Data group of combination of class attribute and frequency value of measurement value, and data group of frequency component and spectrum value And a combination data group of the spectrum value class attribute and the frequency value, a combination data group of the measurement value class attribute and the frequency value regarding the observation data, a combination data group of the frequency component and the spectrum value, and A combination data group of spectral value class attributes and frequency values is prepared, and further, the first determination for each of the three types of combination data group for each of the reference data and the observation data by Equation 1 and Equation 2 Means for calculating an index value, means for calculating a second determination index value for each of the three types of combination data groups with respect to each of the reference data and the observation data by Equation 3 and Equation 4, and reference data by Equation 5 Means for calculating a third determination index value for each type of the three types of combination data group for each of the observation data and the observation data; Integration by action mode using Equations 6 to 8 using at least two decision index values of the first to third decision index values for each type of the three types of combination data groups relating to the quasi-data and the observation data. Means for calculating the judgment value and means for identifying the type of human behavior in the behavior identification target time zone based on the integrated judgment value for each behavior mode are provided.

  Further, the human behavior identification program of the present invention includes a process for creating a frequency distribution of measurement values of reference data acquired as a time history waveform by measurement related to operation information for each behavior mode, and a process for Fourier transforming the reference data And frequency distribution of observed data acquired as a time history waveform by processing to create spectrum of frequency values for each frequency component after Fourier transform of reference data and measurement of motion information in the action identification target time zone , A process for Fourier transforming the observation data, and a process for creating a frequency distribution of the spectrum values for each frequency component after the Fourier transform of the observation data. Combined data group, frequency component and spectrum of class attribute and frequency value for each mode Combination data group, combination data group of spectral value class attribute and frequency value, and combination data group of measurement value class attribute and frequency value regarding observation data, frequency component and spectral value And a combination data group of spectral value class attributes and frequency values, and further, three types of combination data groups for each of the reference data and the observation data according to Equation 1 and Equation 2. A process for calculating a first determination index value for each, a process for calculating a second determination index value for each type of the three types of combination data groups with respect to each of the reference data and the observation data by Formula 3 and Formula 4; The third determination for each type of the three types of combination data group with respect to each of the reference data and the observation data by Equation 5 Expression 6 using at least two determination index values among the first to third determination index values for each type of the three types of combination data groups relating to the standard value calculation process and the reference data and the observation data Or calculating the integrated judgment value for each behavior mode by Formula 8 and causing the computer to perform processing for identifying the type of human behavior in the behavior identification target time zone based on the integrated judgment value for each behavior mode. Yes.

Where Χ: evaluation value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.

Where P: first determination index value,
χ ₁ ² (Χ _{d, j} ): represents a chi-square distribution value with one degree of freedom corresponding to Χ _{d, j} .

Where, B: second determination index value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
However, Equation 4 holds for N (Ba _{d, j, k} ) and N (Ob _{d, k} ).

Where R: third determination index value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
Moreover, Ba _{d, j} ¯ (correctly, Ba _d, on the _j - stick) is an average value for all of the k of _{Ba d, j, k, Ob} d ¯ ( correctly, on the Ob _d (-) Is an average value for all _k of Ob _{d, k} .

Where Pr: integrated judgment value,
I: first determination index value, second determination index value, or third determination index value,
C: weighting factor,
m: identifier indicating the type of determination index value,
j: identifier indicating the type of action mode,
Ni: Represents the number of types of determination index values used.
Further, I _m ￣ (correctly, “-” is added on I _m ) is an average value for all _j of I _{j, m} .
However, C is determined so as to maximize Sp ² represented by Expression 8 using Expression 7 as a constraint.

Where Sp ^{2 is the} amount of dispersion,
no: represents the number of action modes.
In addition, Pr 正 し く (correctly, “-” is attached on Pr) is an average value for all _j of Pr _j .

  Therefore, according to the identification method, identification apparatus, and identification program of these human actions, the time information, phase information, or frequency characteristic information of the reference data and observation data acquired as a time history waveform is removed and used. Therefore, identification processing of human motion / behavior is performed by data in which fluctuation of human motion / behavior in the measurement data is absorbed.

  According to these human behavior identification methods, identification devices, and identification programs, the human behavior / behavior identification process is performed by reference data and observation data collected by only one measuring device. Also good.

  According to the identification method, identification device, and identification program for these human actions, out of the three types of determination index values for each of the three types of combination data groups (that is, nine determination index values in total) Since at least two of them are used to calculate the integrated judgment value for evaluating the degree of similarity between the reference data and the observation data, the degree of similarity is evaluated in a multifaceted and comprehensive manner. The operation / behavior identification process is performed, and the human action / behavior identification process can be performed by an appropriate combination of the determination index values according to the characteristics of the identification target person and the type of action.

  The human behavior identification method, identification apparatus, and identification program according to the present invention are any one of vibration displacement data, vibration speed data, vibration acceleration data, and strain data acquired as a time history waveform. May be used as reference data and observation data. In this case, the reference data and the observation data are collected without requiring particularly expensive measuring equipment or advanced technology.

  According to the human behavior identification method, identification apparatus, and identification program of the present invention, human motion / behavior identification processing can be performed using data in which fluctuations of human motion / behavior in measurement data are absorbed. It becomes possible to identify human movements and actions with high accuracy.

  According to the human behavior identification method, identification apparatus, and identification program of the present invention, the human motion / behavior identification process is performed based on reference data and observation data collected by only one measuring device. Therefore, it is possible to reduce the construction cost of the mechanism according to the present invention, and it is possible to improve versatility.

  According to the identification method, identification apparatus, and identification program for human behavior of the present invention, it is also possible to evaluate human behavior and behavior by evaluating the degree of similarity in a multifaceted and comprehensive manner, and In addition, it is possible to perform human action / behavior identification processing with an appropriate combination of judgment index values according to the characteristics of the person being identified and the type of action, further improving the accuracy of human action / behavior identification It becomes possible to make it.

  The human behavior identification method, identification apparatus, and identification program according to the present invention are such that any one of vibration displacement, velocity, or acceleration data or strain data is used as reference data and observation data. In that case, since it is possible to collect the reference data and observation data without the need for particularly expensive measurement equipment and advanced technology, it becomes possible to reduce the construction cost of the mechanism according to the present invention, As a result, it is possible to improve versatility.

It is a flowchart which shows an example of embodiment of the identification method of the human action of this invention. It is a functional block diagram of a human action identification device realized by the program when the human action identification method of the embodiment is implemented using a human action identification program. It is a figure which shows the time history waveform obtained by measurement of experiment. It is a figure which shows the spectrum intensity distribution obtained by Fourier-transforming the time history waveform of FIG. It is a figure which shows frequency distribution of the amplitude value of the time history waveform of FIG. It is a figure which shows frequency distribution of the spectrum value of the spectrum intensity distribution of FIG.

  Hereinafter, the configuration of the present invention will be described in detail based on an example of an embodiment shown in the drawings.

  In the following description, in order to clarify that it is a unit, a symbol or character as a unit may be enclosed in [].

  FIG. 1 and FIG. 2 show an example of embodiments of a human behavior identification method, identification apparatus, and identification program of the present invention.

  In the human behavior identification method according to the present embodiment, a frequency distribution of measurement values of reference data acquired as a time history waveform by measurement related to operation information for each behavior mode (step A-1) is created (step A-2). The reference data is subjected to Fourier transform (step A-3), and a frequency distribution of spectrum values for each frequency component after the Fourier transform is created (step A-4), and the operation information in the action identification target time zone (Step B-1) The frequency distribution of the measurement values of the observation data acquired as a time history waveform is created (Step B-2) and the observation data is Fourier transformed (Step B-3). A frequency distribution of spectrum values for each frequency component after conversion is created (step B-4), and thereby, an action mode related to reference data Data group of combination of frequency attribute and frequency value, data group of frequency component and spectral value, data group of spectral value class and frequency value, and observation data A combination data group of the attribute of the measurement value class and the frequency value, a combination data group of the frequency component and the spectrum value, and a combination data group of the attribute of the spectrum value and the frequency value are prepared, and these reference data And the first judgment index value (Step B-5) calculated by Formula 9 and Formula 10, and Formula 11 and Formula 12 are calculated for each of the three types of combination data groups related to the observation data. The second determination index value (step B-6) and the third determination index value (step B-7) calculated by Equation 13 Are used to calculate an integrated determination value for each behavior mode using Equations 14 to 16 (step B-8), and based on the integrated determination value for each behavior mode, in the action identification target time zone. The type of human action is identified (step B-9) (see FIG. 1).

Where Χ: evaluation value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.

Where P: first determination index value,
χ ₁ ² (Χ _{d, j} ): represents a chi-square distribution value with one degree of freedom corresponding to Χ _{d, j} .

Where, B: second determination index value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
However, Equation 12 holds for N (Ba _{d, j, k} ) and N (Ob _{d, k} ).

Where R: third determination index value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
Moreover, Ba _{d, j} ¯ (correctly, Ba _d, on the _j - stick) is an average value for all of the k of _{Ba d, j, k, Ob} d ¯ ( correctly, on the Ob _d (-) Is an average value for all _k of Ob _{d, k} .

Where Pr: integrated judgment value,
I: first determination index value, second determination index value, or third determination index value,
C: weighting factor,
m: identifier indicating the type of determination index value,
j: identifier indicating the type of action mode,
Ni: Represents the number of types of determination index values used.
Further, I _m ￣ (correctly, “-” is added on I _m ) is an average value for all _j of I _{j, m} .
However, C is determined so as to maximize Sp ² represented by Expression 16 using Expression 15 as a constraint.

Where Sp ^{2 is the} amount of dispersion,
no: represents the number of action modes.
In addition, Pr 正 し く (correctly, “-” is attached on Pr) is an average value for all _j of Pr _j .

  The human behavior identification method may be implemented by a human behavior identification device. The human behavior identification device 10 according to the present embodiment acquires reference data acquired as a time history waveform by measurement related to operation information for each behavior mode, and acquired as a time history waveform by measurement related to motion information in the action identification target time zone. A data receiving unit 11a as a means for receiving the input of the observed data, a measured value frequency distribution creating unit 11b as a means for creating a frequency distribution of the measured values of the reference data and the observed data, and the reference data and the observed data as Fourier A Fourier transform section 11c as means for transforming, and a spectrum value frequency distribution creation section 11d as means for creating a frequency distribution of spectrum values for each frequency component after Fourier transform of the reference data and observation data. The combination of the class attribute and the frequency value of the measurement value for each behavior mode related to the reference data Data group, combination data group of frequency component and spectrum value, combination data group of spectrum value class attribute and frequency value, and combination of measurement class attribute and frequency value of observation data A data group, a combination data group of frequency components and spectral values, and a combination data group of spectral value class attributes and frequency values are prepared, and further, each of the reference data and the observation data is expressed by Equation 9 and Equation 10, respectively. A chi-two multiplication determining unit 11e as a means for calculating a first determination index value for each type of the three types of combination data groups, and three types of combinations of the reference data and the observation data according to Equations 11 and 12 A virtual calculator 11f as a means for calculating a second determination index value for each type of data group, 3 is a correlation coefficient calculation unit 11g as a means for calculating a third determination index value for each type of the three types of combination data group relating to each of the reference data and the observation data, and each of the reference data and the observation data Integration as means for calculating an integrated determination value for each action mode according to Expressions 14 to 16 using at least two determination index values of the first to third determination index values for each of the three types of combination data groups The determination value calculation unit 11h and an action identification unit 11i as a means for identifying the type of human action in the action identification target time zone based on the integrated determination value for each action mode.

  Further, the human behavior identification method and the human behavior identification device may be implemented and realized by executing a human behavior identification program on a computer. Here, a case will be described in which a human behavior identification device 17 is realized by executing the human behavior identification program 17 on the computer 10 and a human behavior identification method is implemented.

  FIG. 2 shows an overall configuration of a computer 10 (which is also a human behavior identification device 10 in the present embodiment) for executing the human behavior identification program 17 of the present embodiment. The computer 10 (human action identification device 10) includes a control unit 11, a storage unit 12, an input unit 13, a display unit 14, and a memory 15, which are connected to each other by a signal line such as a bus.

  The control unit 11 performs a calculation related to the control of the entire computer 10 and the identification of the human behavior by the human behavior identification program 17 stored in the storage unit 12, and is, for example, a CPU (Central Processing Unit). is there.

  The storage unit 12 is a device that can store at least data and programs, and is, for example, a hard disk.

  The input unit 13 is an interface (that is, an information input mechanism) for giving at least an operator's command and various information to the control unit 11, and is, for example, a keyboard or a mouse. For example, a plurality of types of interfaces such as a keyboard and a mouse may be provided as the input unit 13.

  The display unit 14 performs drawing / display of characters, figures, images, and the like under the control of the control unit 11 and is, for example, a display.

  The memory 15 serves as a memory space that is a work area when the control unit 11 executes various controls and operations, and is, for example, a RAM (abbreviation of Random Access Memory).

  Then, the control unit 11 of the computer 10 (denoted as “human action identification device 10”) executes the human action identification program 17 and thereby measures the time history by measuring the operation information for each action mode. Reference data acquired as a waveform, and a data receiving unit 11a that performs a process of receiving input of observation data acquired as a time history waveform by measurement related to motion information in the action identification target time zone, and measurement of reference data and observation data A measured value frequency distribution creating unit 11b that performs a process of creating a frequency distribution of values, a Fourier transform unit 11c that performs a process of Fourier transforming the reference data and the observation data, and each frequency component after the Fourier transform of the reference data and the observation data And a spectral value frequency distribution creating unit 11d that performs processing for creating a frequency distribution of the spectral values of As a result, the combination data group of the class attribute of the measurement value and the frequency value for each behavior mode regarding the reference data, the combination data group of the frequency component and the spectrum value, and the class attribute of the spectrum value and the frequency value Combination data group, combination data group of measurement value class attribute and frequency value for observation data, combination data group of frequency component and spectrum value, and spectrum value class attribute and frequency value A combination data group is prepared. When the human behavior identification program 17 is executed in the control unit 11 of the human behavior identification device 10, three types of combination data relating to each of the reference data and the observation data are further represented by Equation 9 and Equation 10. A chi-two multiplication determining unit 11e that performs processing for calculating a first determination index value for each type of group, and three types of combination data groups for each of the reference data and the observation data according to Equations 11 and 12 for each type. A process for calculating a third determination index value for each type of the three types of combination data groups with respect to each of the reference data and the observation data according to Equation 13 by the batcharia calculation unit 11f that performs the process of calculating the second determination index value. For each type of the combination data group of the three types relating to the correlation coefficient calculation unit 11g for performing the reference data and the observation data An integrated determination value calculation unit 11h that performs processing for calculating an integrated determination value for each action mode according to Expressions 14 to 16 using at least two determination index values of the first to third determination index values; A behavior identification unit 11i that performs processing for identifying the type of human behavior in the behavior identification target time zone based on the integrated determination value is configured.

  The human behavior identification method of the present invention generally includes a stage in which reference data is prepared (phase A) and a stage in which the type of human behavior in a facility such as a residence is identified (phase B). .

  Then, as the execution of the human behavior identification method, reference data is first prepared (phase A). Specifically, first, measurement of operation information is performed for each preset behavior mode, and reference data is obtained. Measurement data is acquired (step A-1).

  The reference data is used in the preparation of reference data (also referred to as teacher data or reference data) used in the stage (phase B) in which the type of human behavior is identified, and identification within the target facility Collected and acquired for each type of action / behavior that the target person is supposed to take in the facility (referred to as “behavior mode”).

  The behavior mode is not limited to a specific type of movement or behavior, but is relatively frequently performed as an identification target person's movement or behavior in the targeted facility or identified in the targeted facility. Appropriate ones are appropriately selected and set in consideration of things that need to be grasped as the actions and actions of the person or the purpose of identifying the actions of the person to be identified.

  As specific examples of behavior modes, the following are just examples: stationary state, indoor movement (in other words, walking state), sitting / leaving in a chair, walking in a corridor, elevating stairs, using a vacuum cleaner, Examples include the use of toilets, the use of baths, and the state of lying on the floor and turning over repeatedly.

  If necessary, the behavior mode may be set for each floor when the facility has a multi-floor structure, or the behavior mode may be set for each room when the facility has a plurality of rooms. .

  Further, the reference data is prepared using data (that is, reference data) when the identification target person actually performs the operation / behavior for each behavior mode in the target facility. For this reason, when there are a plurality of identification target persons in the target facility, the reference data is acquired for each identification target person and the reference data is prepared for each identification target person.

  In the present invention, information relating to a phenomenon that is generated when a person moves and that can be measured as a time history waveform is used as the operation information. That is, the operation information in the present invention is not limited to specific information as long as it is information related to the phenomenon as described above.

  As the operation information in the present invention, specifically, for example, vibration or strain generated in a facility when a person operates can be used. When vibration is used as the operation information, vibration displacement data, vibration speed data, or vibration acceleration data is collected and acquired as a time history waveform by measurement. When strain is used as operation information, strain (particularly dynamic strain) data is collected and acquired as a time history waveform.

  The method of measuring the operation information is specified if, for example, one of the above data is measured and the measured value of any of the above data is output as an electrical signal at a predetermined time, for example. The method and device are not limited to those described above, and an appropriate method or device is appropriately selected in consideration of the situation of the target facility.

  For example, a vibration sensor (specifically, a vibration displacement detection type sensor, a speed detection type sensor, or an acceleration detection type sensor) or a strain sensor is installed in a target facility such as a residence, and measurement related to operation information is performed. Values are collected and acquired.

  Specifically, as an example, vibration speed is measured at a predetermined time interval (in other words, sampling cycle), and the vibration speed value at the predetermined time interval is collected as a measurement value.・ Acquired.

  The measurement time interval [seconds] as the above-mentioned “predetermined time interval (sampling period)” in the measurement of motion information for collecting the reference data and the observation data as a time history waveform is set to a specific time [second]. It is not limited, and is appropriately set to an appropriate value in consideration of the specification of the measuring device used for measurement. Specifically, for example, as an example only, an appropriate value can be set in the range of about 1 to 100 milliseconds. Note that the sampling period in the range of about 1 to 100 milliseconds means that the sampling frequency is 10 to 1000 [Hz].

  At least one measuring device for measuring the operation information is installed in the target facility.

  In order to be able to properly capture the movements and behaviors of people at the target facility, the measurement equipment can be used in places and places where the phenomenon generated by the movement of the person is thought to be propagated or manifested as much as possible. It is preferable to be installed. Specifically, for example, it is preferable that the measuring device is installed on the floor of the largest room / space or the floor of the hallway in the target facility.

  When one measuring device is installed in the target facility, it is preferable that the measuring device is installed in a place / location where the behavior of a person in the entire target facility can be grasped by the one measuring device. In addition, when multiple measuring devices are installed in the target facility, the human behavior in the entire target facility can be efficiently grasped by these multiple measuring devices (in other words, the sensing range of each measuring device is necessary. It is preferable that the measuring device is installed at a place / location where the entire facility can be covered without omission as a sum of the sensing ranges of the measuring devices without overlapping.

  Measurements related to operation information for collecting and acquiring reference data are integrated as data from the start to the end of the action mode for each action mode (in other words, from the start to the end of the action mode as a continuous target) ) Done to be measured.

  The measurement related to the operation information for each behavior mode, that is, the measurement of the reference data may be performed in the same measurement time [second] for all the behavior mode types, or as a series of behaviors for each behavior type. The measurement may be performed at different measurement time [seconds] depending on the type of action mode, for example, considering the time to complete.

  Note that the measurement time [seconds] related to the operation information for each behavior mode is not limited to a specific time length [seconds], and is appropriate considering the time required to complete the selected and set behavior mode. The time length is appropriately set. Specifically, for example, as an example only, an appropriate time length can be set in a range of about 30 to 90 seconds.

  And the measured value regarding the movement information for every action mode acquired by measurement is input into the identification apparatus 10 of a human action via the data receiving part 11a.

  The measurement value related to the motion information may be input to the human behavior identification device 10 via, for example, a data server, or the human behavior identification device 10 and the measurement device may receive data, control commands, or the like. These signals may be electrically connected so that they can be transmitted and received (ie, input / output).

  When the measurement value is input via the data server, the data server is connected to the human action identification device 10 via a signal line such as a bus, and the measurement value output from the measurement device is transferred to the data server as a data file. Etc., and the measurement values stored as the data file or the like may be read.

  In addition, when a human behavior identification device 10 and a measurement device are electrically connected and a measurement value is input, for example, wired signal transmission / reception using cables connected to each other is used. Signals may be transmitted / received via a wireless signal transmission / reception mechanism using a wireless signal transmitter / receiver connected to each other so that signals can be transmitted / received via the mechanism. It is also possible to be electrically connected so as to be possible. Then, the measurement values may be input by these signal transmission / reception mechanisms.

  Furthermore, a measurement value may be input to the human action identification device 10 by using a data server or a wired / wireless signal transmission / reception mechanism in combination.

  The measurement value is shaped by a device / mechanism such as a high-pass filter or a low-pass filter, taking into account the specifications / performance of the measurement device used for the measurement, for example, if necessary. May be input.

  In addition, the time when each measured value was measured was output together with the measured value by the clock function provided in the measuring device and recorded in association with the measured value, or output from the measuring device at the same time as the measurement. The time when the measurement value is recorded or input to the data server or the human behavior identification device 10 is recorded in association with the measurement value.

  In the description of the present invention, the time recorded in association with each measurement value is referred to as “measurement time”, and the combination data of the measurement time and the measurement value at the measurement time is referred to as “measurement data”. Call. The measurement data as the combination data is time history waveform data such as a time history vibration waveform (specifically, a time history displacement waveform, a time history speed waveform, or a time history acceleration waveform) or a time history distortion. It is a waveform.

  Then, the measurement value is associated with the measurement time for each behavior mode by the data receiving unit 11a (in other words, the combination of the behavior mode type, the measurement time, and the measurement value) is stored in the memory 15.

  Subsequently, the measurement value frequency distribution creating unit 11b of the control unit 11 creates the frequency distribution of the measurement values by using the measurement data as the reference data measured and acquired by the process of Step A-1. Step A-2).

  Specifically, the measurement value frequency distribution creating unit 11b reads the measurement data stored in the memory 15 in the process of step A-1 for each behavior mode, and a frequency distribution (histogram) focusing on the absolute value of the measurement value. Is created.

  For example, when the motion information (in other words, the measured value) is a vibration displacement, a frequency distribution of the absolute value of the displacement is created, and in the case of the vibration speed, a frequency distribution of the absolute value of the speed is created. In the case of vibration acceleration, a frequency distribution of absolute values of acceleration is created, and in the case of strain, a frequency distribution of absolute values of strain is created.

  By converting the waveform data of the time history measured and acquired by the process of step A-1 into a frequency distribution focusing on the absolute value of the measurement value, time information and frequency characteristic information are obtained from the waveform data of the time history. It is removed and the waveform data of the time history is converted into data focusing only on the amplitude value.

  The width of the measured value (absolute value) class (also called class interval or bin width) when creating the frequency distribution is not limited to a specific value, and the results of prior analysis are taken into account. Known techniques such as the Scott method and the Sturges method are also referred to, and are appropriately set to appropriate values.

  Then, the measured value frequency distribution creating unit 11b associates the frequency value with the class for each behavior mode (in other words, as the combination data of the behavior mode type, class attribute, and frequency value) in the memory 15. Be made.

  Subsequently, the Fourier transform of the measurement data as the reference data measured and acquired by the process of Step A-1 is performed by the Fourier transform unit 11c of the control unit 11 (Step A-3).

  Specifically, the Fourier transform unit 11c reads the measurement data stored in the memory 15 in the process of step A-1 for each behavior mode as time-series waveform data in the order of measurement time, and performs Fourier transform. A spectrum value for each frequency component is calculated.

  In addition, when the motion information (in other words, the measured value) is, for example, a vibration displacement, an absolute displacement spectrum value is calculated, and in the case of a vibration speed, an absolute velocity spectrum value is calculated. Is calculated as an absolute acceleration spectrum value, and in the case of strain, an absolute strain spectrum value is calculated.

  By performing Fourier transform on the waveform data of the time history measured and acquired by the processing of step A-1, phase information is removed from the waveform data of the time history, and the waveform data of the time history is spectrum. Converted into components. As a result, it is expected that similar frequency characteristics such as an increase in the same frequency component are easily captured regardless of the elapsed time when the maximum amplitude appears if the same action has periodicity.

  The frequency component pitch [Hz] when performing the Fourier transform is not limited to a specific value, and is appropriately set to an appropriate value in consideration of the specifications of the measuring instrument used for the measurement and the sampling period. Is done.

  Then, the Fourier transform unit 11c associates the calculated spectrum value with the frequency component for each behavior mode (in other words, as the combination data of the behavior mode type, the frequency component, and the spectrum value) and stores it in the memory 15. It is done.

  Subsequently, the spectrum value frequency distribution creating unit 11d of the control unit 11 uses the spectrum value for each frequency component calculated by the process of step A-3 to create a frequency value frequency distribution (step A-). 4).

  Specifically, the spectrum value frequency distribution creating unit 11d reads the combination data of the frequency component and the spectrum value stored in the memory 15 in the process of step A-3 for each behavior mode, and the frequency focused on the spectrum value. A distribution (histogram) is created.

  By converting the spectrum value data for each frequency component calculated by the process of step A-3 into a frequency distribution focusing on the spectrum value, the frequency characteristic information is removed from the spectrum value data for each frequency component, and the frequency The spectral value data for each component is converted into data focusing only on the amplitude value (spectral value).

  The width of the spectrum value class (also referred to as class interval or bin width) when creating the frequency distribution is not limited to a specific value. For example, Scott's method or A known value such as the Sturges method is also referred to, and the value is appropriately set to an appropriate value.

  Then, the spectrum value frequency distribution creating unit 11d associates the frequency value with the class for each action mode (in other words, as combination data of the action mode type, class attribute, and frequency value) in the memory 15. Be made.

  Through the processes in steps A-1 to A-4, the combination data of the class attribute and the frequency value focusing on the absolute value of the measured value for each behavior mode (reference value frequency distribution data) ), Combination data of frequency components and spectrum values (referred to as “frequency spectrum value data”), and combination data of class attributes and frequency values focusing on spectrum values (“spectrum value frequency distribution data” and Called) is maintained.

  Next, the type of action of the person in the facility such as a residence is identified (phase B).

  According to the present invention, identification of the type of human behavior (behavior mode) is performed in a time zone (referred to as “behavior identification target time zone”) that is a target for identifying behavior in the target facility. Acquired measurement data (referred to as “observation data”) is used.

  For this reason, the measurement regarding the operation information in the action identification target time zone is performed, and the measurement data as the observation data is acquired (step B-1).

  The measurement related to the operation information for identifying the type of human action is performed under basically the same conditions as the measurement in the process of Step A-1 of Phase A for the preparation of reference data.

  Then, the measurement value related to the operation information in the action identification target time zone obtained by the measurement is input to the human behavior identification device 10 via the data receiving unit 11a, and is associated with the measurement time by the data receiving unit 11a. (In other words, it is stored in the memory 15 as combination data of measurement time and measurement value).

  Here, in the process of identifying the type of human action (phase B), the measurement data of the same measurement time [seconds] as the measurement related to the operation information for each action mode in the preparation of the reference data (phase A) is used.

  Therefore, when the measurement of the reference data in phase A is performed in the same measurement time [seconds] for all types of behavior modes, the measurement of the observation data in phase B is the same as the measurement of the reference data in phase A. It is performed with the same measurement time [second]. However, measurement is performed for a time longer than the measurement time of the reference data in phase A, and data having the same time length as that of phase A is extracted by an appropriate method and used as observation data for phase B. May be.

  On the other hand, when the measurement of the reference data in phase A is performed at different measurement times [seconds] depending on the type of action mode, the observation data in phase B is measured in the measurement time of reference data in phase A [seconds]. The measurement time [seconds] is the same as or longer than the longest measurement time. Then, a plurality of time length data corresponding to different measurement times depending on the type of the behavior mode of phase A is extracted by an appropriate method and used as observation data of phase B.

  Further, the identification process (phase B) of the type of human action may be performed on, for example, measurement data that has been measured and stored in advance, or at a predetermined interval or some event, etc. As a trigger, measurement data repeatedly measured may be immediately and repeatedly performed. That is, there is no particular restriction on how to set the action identification target time zone in the present invention, in other words, any time zone in which measurement data with a length longer than the measurement time of the reference data is acquired becomes the action identification target time zone. obtain. Therefore, the method of setting the action identification target time zone in the present invention is arbitrary, and acquisition of observation data can be performed at any time and in a timely manner.

  In the case of performing iteratively, specifically, for example, the start of a process is triggered by the arrival of a predetermined time interval, or the process starts when measurement data is input to the human action identification device 10 The process of phase B may be performed by using as a trigger.

  Subsequently, the measurement value frequency distribution creation unit 11b of the control unit 11 creates the frequency distribution of the measurement values using the measurement data as the observation data measured and acquired by the process of Step B-1. Step B-2).

  Specifically, the measurement value frequency distribution creating unit 11b reads the measurement data stored in the memory 15 in the process of step B-1, and creates a frequency distribution (histogram) focusing on the absolute value of the measurement value. .

  The measurement value frequency distribution for identifying the type of human action is created under the same conditions as the measurement value frequency distribution in the process of step A-2 of phase A for the preparation of reference data.

  Then, the measured value frequency distribution creating unit 11b associates the frequency value with the class (in other words, as combined data of the class attribute and the frequency value) and stores it in the memory 15.

  Subsequently, the Fourier transform of the measurement data as the observation data measured and acquired by the process of Step B-1 is performed by the Fourier transform unit 11c of the control unit 11 (Step B-3).

  Specifically, the measurement data stored in the memory 15 in the process of Step B-1 is read as time-series waveform data in the order of measurement time by the Fourier transform unit 11c, and Fourier transform is performed for each frequency component. A spectral value is calculated.

  The Fourier transform of the measurement data for identifying the type of human action is performed under the same conditions as the Fourier transform in the process of Step A-3 of Phase A for preparing reference data.

  Then, the calculated spectral value is associated with the frequency component by the Fourier transform unit 11c (in other words, as combined data of the frequency component and the spectral value) and stored in the memory 15.

  Subsequently, the spectrum value frequency distribution creating unit 11d of the control unit 11 creates a frequency distribution of spectrum values using the spectrum value for each frequency component calculated by the process of step B-3 (step B-). 4).

  Specifically, the spectrum value frequency distribution creating unit 11d reads combination data of frequency components and spectrum values stored in the memory 15 in the process of step B-3, and a frequency distribution (histogram) focusing on the spectrum values. Is created.

  The creation of the frequency distribution of the spectrum values for identifying the type of human action is performed under the same conditions as the frequency distribution of the histogram values in the process of Step A-4 of Phase A for the preparation of reference data.

  Then, the spectrum value frequency distribution creating unit 11d associates the frequency value with the class (in other words, as combined data of the class attribute and the frequency value) and stores it in the memory 15.

  Through the processing of steps B-1 to B-4 described above, the combination data of the class attribute focusing on the absolute value of the measurement value and the frequency value related to the observation data (that is, “measurement value frequency distribution data”), the frequency component And spectrum value combination data (ie, “frequency spectrum value data”), and combination data of class attributes and frequency values focusing on spectrum values (ie, “spectrum value frequency distribution data”) are prepared.

  In the present invention, the type of action of the person when the observation data is measured is identified based on whether the observation data is similar to the data of any action mode of the reference data for each action mode.

  Therefore, in the present invention, three types of combination data for each behavior mode created in the processing of steps A-2 to A-4 using the reference data for each behavior mode and the observation data are used for step B. -2 to B-4 are used in combination with three types of combination data, and an index (referred to as a “determination index”) for evaluating the degree of similarity of observation data with reference data of each behavior mode A plurality of values (referred to as “determination index values”) are calculated.

  For the calculation of the determination index value, first, the chi-multiplication constant unit 11e of the control unit 11 uses the data obtained by the processing of steps A-2 to A-4 and steps B-2 to B-4. Calculation of the similarity determination value applying the square distribution is performed (step B-5).

  Specifically, the combination data stored in the memory 15 in the processing of steps A-2 to A-4 (specifically, the measured value frequency distribution data for each action mode related to the reference data) by the chi-two multiplication determining unit 11e. , Frequency spectrum value data, and spectrum value frequency distribution data), and combination data stored in the memory 15 in the processing of steps B-2 to B-4 (specifically, measured value frequency distribution data regarding observation data, Frequency spectrum value data and spectrum value frequency distribution data) are used to calculate an evaluation value し た by applying a chi-square value representing the degree of matching between the reference data of each behavior mode and the observation data, using Equation 17. The

Here, ：: Evaluation value applying chi-square value,
Ba: Value created from reference data,
Ob: Value created from observation data,
d: identifier indicating the type of combination data,
j: identifier indicating the type of action mode,
k: identifier indicating the type of component,
n (d): Represents the total number of components by type (d) of the combination data group.

  The identifier d indicating the type of combination data is, for example, d = 1 for the measured value frequency distribution data group, d = 2 for the frequency spectrum value data group, and d = 3 for the spectrum value frequency distribution data group. It is given as follows.

  An identifier j representing the type of behavior mode is given to each behavior mode selected and set in phase A. For example, when there are 15 types of action modes set in phase A, j = 1, 2, 3,.

  The value Ba created from the reference data and the value Ob created from the observation data are frequency values for the measured value frequency distribution data and the spectral value frequency distribution data, and are spectrum values for the frequency spectrum value data.

  The identifier k representing the component type is given as an identifier representing the class attribute type for the measured value frequency distribution data and the spectral value frequency distribution data, and the identifier representing the frequency component type for the frequency spectrum value data. (Specifically, k = 1, 2, 3,...)

  The total number n (d) of components by type (d) of the combination data group is the total number of class attribute types for the measured value frequency distribution data and the spectral value frequency distribution data, and the frequency component for the frequency spectrum value data. The total number of types. Therefore, k = 1, 2, 3,..., N (d).

Further, the evaluation value Χ _{d, j obtained} by applying the chi-square value calculated by the above equation 17 is used by the chi-2 multiplication determining unit 11e, and the reference data and the observation data of each action mode are calculated by the equation 18. A similarity determination value P representing the similarity is calculated.

Where P: similarity determination value applying chi-square distribution,
χ ₁ ² (Χ _{d, j} ): represents a chi-square distribution value with one degree of freedom corresponding to Χ _{d, j} .

The similarity determination value P _{d, j} calculated by Expression 18 is a value in the range of 0 ≦ P _{d, j} ≦ 1, and when it is 1, it means that the similarity is the highest.

Then, the chi-two multiplication determining unit 11e associates the value of the similarity determination value P _{d, j} applying the chi-square distribution as the first determination index value with the type of combination data and the type of action mode. (In other words, it is stored in the memory 15 as combination data of the combination data type d, the action mode type j, and the similarity determination value P _{d, j} ).

  Subsequently, the similarity determination value using the data obtained by the processes of Steps A-2 to A-4 and Steps B-2 to B-4 by the virtual calculation unit 11f of the control unit 11 and applying the virtual coefficient. Is calculated (step B-6).

  Specifically, the combination data stored in the memory 15 in the processing of Steps A-2 to A-4 and the memory 15 stored in the processing of Steps B-2 to B-4 by the Batacharia calculating unit 11f. The combination data is used to calculate a similarity determination value B to which a virtual coefficient representing the degree of similarity between the reference data for each behavior mode and the observation data is applied according to Equation 19.

Here, B: represents a similarity determination value using a batcharia coefficient.

  Ba, Ob, d, j, k, and n (d) in Equation 19 are the same as those in Equation 17.

However, as N (Ba _{d, j, k} ) in Expression 19, the value Ba created from the reference data is normalized according to the combination data type (d) and action mode type (j) by Expression 20. A value is used, and N (Ob _{d, k} ) is a value obtained by normalizing the value Ob generated from the observation data for each type (d) of the combination data according to Equation 20.

The similarity determination value B _{d, j} calculated by Equation 19 is a value in the range of 0 ≦ B _{d, j} ≦ 1, and when it is 1, it means that the similarity is the highest.

Then, the virtual value calculation unit 11f associates the value of the similarity determination value B _{d, j} applying the virtual parameter as the second determination index value with the type of combination data and the type of action mode (in other words, (As combination data of combination data type d, action mode type j, and similarity determination value B _{d, j} ).

  Subsequently, the correlation coefficient calculation unit 11g of the control unit 11 calculates the correlation coefficient using the data obtained by the processes of steps A-2 to A-4 and steps B-2 to B-4. (Step B-7).

  Specifically, the correlation coefficient calculation unit 11g stores the combination data stored in the memory 15 in the processing of steps A-2 to A-4 and the memory 15 in the processing of steps B-2 to B-4. Using the combined data, a correlation coefficient R representing the degree of association between the reference data for each behavior mode and the observation data is calculated using Equation 21.

Here, R: correlation coefficient.

  Ba, Ob, d, j, k, and n (d) in Equation 21 are the same as those in Equation 17.

Ba _{d, j} ￣ (correctly, “-” is added on top of Ba _{d, j} ) is the combination data type (d) and action mode type of the value Ba _{d, j, k} created from the reference data. (j) It is an average value regarding the type (k) of all other components.

In addition, Ob _d ￣ (correctly, “-” is added to Ob _d ) is the value of all the components (k) of the value Ob _{d, k} created from the observation data by type (d) of the combination data. ) Is the average value.

The correlation coefficient R _{d, j} calculated by Equation 21 is a value in the range of −1 ≦ R _{d, j} ≦ 1, and when it is 1, it means that the similarity is the highest.

Then, the correlation coefficient calculation unit 11g associates the value of the correlation coefficient R _{d, j} with the type of combination data and the type of action mode as the third determination index value (in other words, the type of combination data d, action mode type j, and correlation coefficient R _{d, j} as combination data).

Through the processing of steps B-5 to B-7, three types of combination data (specifically, three types of measurement value frequency distribution data, frequency spectrum value data, and spectrum value frequency distribution data) and determination index values are obtained. For each of the nine types of determination indexes defined as combinations with the three types, the determination index values of observation data for each behavior mode (specifically, P _{d, j} , B _{d, j} , and R _{d, j} ) Is calculated.

  Next, the integrated determination value calculation unit 11h of the control unit 11 calculates the integrated determination value for each action mode using the determination index values obtained by the processing of Steps B-5 to B-7 (Step S5). B-8).

Specifically, the integrated determination value calculation unit 11h performs combination data of the combination data type d, action mode type j, and similarity determination value _{Pd, j} stored in the memory 15 in the process of step B-5. The combination data of the combination data type d, the action mode type j, and the similarity determination value B _{d, j} stored in the memory 15 in the process of Step B-6, and the memory 15 in the process of Step B-7. The combination data of the stored combination data type d, action mode type j, and correlation coefficient R _{d, j} is used, and the reference data for each action mode is expressed by Equation 22 using the principal component analysis methodology. An integrated judgment value Pr _j for comprehensively representing the degree of similarity with the observation data is calculated.

Where Pr: integrated judgment value,
I: Determination index value,
C: weighting factor,
m: identifier indicating the type of judgment index,
j: identifier indicating the type of action mode,
Ni: Number of types of determination index used (where Ni = 2, 3, 4,..., 9)
Respectively.

The determination index value I _{j, m} is specifically the determination index value of the observation data with respect to the reference data of a certain behavior mode, for each type (d) of the combination data (specifically, the measured value frequency distribution data , Frequency spectrum value data, and spectrum value frequency distribution data) and the similarity determination value P _{d, j} as the first determination index value for each behavior mode type (j), the similarity as the second determination index value This corresponds to the degree determination value B _{d, j} and the correlation coefficient R _{d, j} as the third determination index value.

I _m ￣ (precisely, “-” is added to I _m ) indicates all behaviors of the judgment index value I _{j, m} of the observation data with respect to the reference data of each behavior mode according to the judgment index type (m). It is an average value regarding the mode type (j).

The weighting factor C is determined by the Lagrange's undetermined coefficient method so that the variance Sp ² of the integrated decision value expressed by the equation 24 is maximized using the equation 23 as a constraint condition, and the classification index type (m) It is determined as a combination of the value C _m for each.

Where Sp ^{2 is the} amount of dispersion,
no: represents the number of action modes.

Pr ￣ (correctly, “-” is attached on Pr) is an average value of all the behavior mode types (j) of the integrated determination value Pr _j for the observation data with respect to the reference data of each behavior mode.

In general, the same number of weighting factors C determined as combinations of the values C _m for each determination index type (m) are obtained as many as the number of determination index types (m). Therefore, the integrated determination value Pr _j is calculated for each behavior mode type (j) by the same number as the number of determination index types (m) (that is, Ni) by Equation 22.

Then, among the plurality of integrated determination values Pr _j , a value calculated by a combination of the weighting factor C whose magnitude of the dispersion amount Sp ² represented by Expression 24 is the xth is an integrated determination value Pr of the x-th principal component. _j .

Further, the integrated judgment value Pr _j for each behavior mode type (j) calculated by Equation 23 is any combination of nine kinds of judgment indices, that is, all combinations for selecting 2 to 9 kinds from among the nine kinds. It is calculated for each of (represented as “combination of determination indices ₉ C _{2 to 9} C ₉ ” using _p C _r , which is a symbol representing the number of combinations obtained by selecting r out of _p ) The

That is, the integrated determination value Pr _j is calculated from the first principal component to the x-th principal component for each combination of determination indices ₉ C _{2 to 9} C ₉ for each behavior mode type (j). Here, x is the number of selected determination indices, and 2 ≦ x ≦ 9.

Then, the integrated determination value calculation unit 11h determines the values of the integrated determination value P1 _j of the first principal component and the integrated determination value P2 _j of the second principal component (mathematically, the principal component score) as the type of action mode ( j) and the types of combinations of the determination indexes ₉ C _{2 to 9} C ₉ (in other words, the types of combinations of the determination indexes ₉ C _{2 to 9} C ₉ , the action mode type j, the first main (As a combination data of the component integrated determination value P1 _j and the second principal component integrated determination value P2 _j ).

  Next, the behavior identification unit 11i of the control unit 11 identifies the type of human behavior using the integrated determination value for each behavior mode calculated by the process of Step B-8 (Step B-9). .

Specifically, the action identification unit 11i causes the combination of the determination indices ₉ C _{2 to 9} C ₉ stored in the memory 15 in the process of Step B-8, the action mode type j, the first principal component Combination data of the integrated determination value P1 _j and the integrated determination value P2 _j of the second principal component is used. First, the determination index ₉ C having the maximum integrated determination value P1 _j of the first principal component is maximized. The combination type _{2 to 9} C ₉ and the action mode type (j) are specified. This type of behavior mode is referred to as a “first principal component behavior mode”.

Subsequently, the value of the integrated determination value P2 _j of the second principal component is the maximum in the combination types ₉ C _{2 to 9} C ₉ of the determination index having the maximum value of the integrated determination value P1 _j of the first principal component. The type (j) of the behavior mode that is This type of behavior mode is referred to as a “second principal component behavior mode”.

  When the first principal component behavior mode and the second principal component behavior mode are the same, the human action / behavior at the measurement time (in other words, the behavior identification target time zone) is the first principal component behavior mode. Identified.

On the other hand, when the first principal component behavior mode and the second principal component behavior mode are different, the Mahalanobis distance D is calculated for each of the first principal component behavior mode and the second principal component behavior mode by Equation 25. The
Where D: Mahalanobis distance,
P1: integrated judgment value of the first principal component,
S _p1 : represents the standard deviation of P1 _j .

As P1 _j , the integrated determination value of the first principal component in the first principal component behavior mode or the integrated determination value of the first principal component in the second principal component behavior mode is used.

P1 _j ￣ (correctly, “-” is attached on P1 _j ; the same applies hereinafter) is the integrated determination value of the first principal component among the integrated determination values for each principal component of the observation data with respect to the reference data of each behavior mode. P1 _j is an average value for each behavior mode type (j) (specifically, an average value related to the first principal component behavior mode or an average value related to the second principal component behavior mode).

The value of P1 _j ￣ is an integrated determination of a plurality of first principal components obtained by performing the processes of Steps B-1 to B-8 a plurality of times after the action mode is specified (in other words, specified). It is calculated in advance as an average of the values P1 _j .

S _P1 is the first principal component of the integrated determination value P1 _j of the main component for each of the integration identifying values for observation data for the reference data of each behavior modes, a type of behavioral modes (j) by the standard deviation (specifically Specifically, the standard deviation regarding the first principal component behavior mode or the standard deviation regarding the second principal component behavior mode).

The value of S _P1 is obtained by performing the processes of Steps B-1 to B-8 a plurality of times after the behavior mode is specified (in other words, specified) together with the calculation of the average value P1 _j上述 described above. It is calculated in advance as the standard deviation of the integrated judgment value P1 _j of the plurality of first principal components obtained.

  Then, the behavior mode having the smaller Mahalanobis distance D among the first principal component behavior mode and the second principal component behavior mode is selected as the candidate behavior mode.

Then, it is determined whether or not the value of the integrated determination value P1 _j of the first principal component in the candidate action mode is within the range of P1 _j ￣ ± α.

Then, when the value of the P1 _j is in the range of P1 _j ¯ ± α is, (in other words, behavior identified target time band) measurement time the operation and behavior of people in is identified as a candidate action mode The

On the other hand, if the value of the P1 _j is not in the range of P1 _j ¯ ± α is (in other words, action-identified time zone) measurement time operation and behavior of the human in the set are selected in Phase A It is determined that it is not one of the action modes.

Note that α used in the above identification / judgment processing is not limited to a specific value, and is appropriately set to an appropriate value in consideration of required identification accuracy. Specifically, for example, as an example only, the value of α can be set to the standard deviation S _P1 .

  In addition, when it is determined that it is not one of the action modes selected and set in Phase A, it is confirmed what the person's action / behavior was specifically at the measurement time. In addition, the action mode may be additionally set, and the observation data may be added to the reference data to further improve the reference data.

  As described above, the identification processing of the type of human behavior for the measurement data input to the human behavior identification device 10 in the processing of Step B-1 is completed. Then, depending on the case (in other words, if necessary), the control unit 11 repeatedly performs the processes of Steps B-1 to B-9 according to the trigger for starting the process.

  According to the human behavior identification method, identification device, and identification program configured as described above, the time information, phase information, or frequency characteristic information of the reference data and observation data acquired as a time history waveform is removed. Because it is used, human movement / behavior identification processing can be performed based on data in which fluctuations in human movement / behavior in measurement data are absorbed, and human movement / behavior can be identified with high accuracy. It becomes possible.

  According to the identification method, identification apparatus, and identification program for human behavior configured as described above, the human motion / behavior identification processing is performed based on the reference data and the observation data collected by only one measuring device. Therefore, it is possible to reduce the construction cost of the mechanism according to the present invention, and it is possible to improve versatility.

  According to the human behavior identification method, identification device, and identification program configured as described above, three types of determination index values for each type of the three types of combination data groups (that is, the determination index values are total) At least two of them are used to calculate the integrated judgment value for evaluating the degree of similarity between the reference data and the observed data. The human motion / behavior identification process can be performed by evaluating the degree, and the human motion / behavior identification process is performed by an appropriate combination of judgment index values according to the characteristics of the target person and the type of the behavior. Thus, it becomes possible to further improve the identification accuracy of human movement / behavior.

  Although the above-described embodiment is an example of a preferred embodiment for carrying out the present invention, the embodiment of the present invention is not limited to the above-described embodiment, and the present invention is not deviated from the gist of the present invention. Various modifications can be made.

For example, in the above-described embodiment, the integrated determination value Pr _j for each type (j) of the action mode is calculated for each combination of the determination indexes ₉ C _{2 to 9} C ₉ , but is not limited thereto. Alternatively, the type of the determination index used for the identification process may be selected in advance, and only the determination index value related to the selected determination index may be calculated. In this case, the type of determination index is selected in the range of 2 to 9 types.

In the above-described embodiment, both the integrated principal determination value P1 _j of the first principal component and the integrated principal determination value P2 _j of the second principal component (mathematically, the principal component score) are calculated. However, the present invention is not limited to this, and only the value of the integrated determination value P1 _j of the first principal component may be calculated and used for the identification process. Then, the type of the behavior mode in which the integrated principal determination value P1 _j of the first principal component is maximized is identified as a human action / behavior at the measurement time (in other words, the behavior identification target time zone). Anyway.

In the above-described embodiment, the value of the integrated determination value P1 _j of the first principal component and the value of the integrated determination value P2 _j of the second principal component are considered independently. However, the present invention is not limited to this. The integrated determination value P1 _j of the first principal component and the integrated determination value P2 _j of the second principal component may be averaged and considered. The action mode type in which the average value of the integrated determination value P1 _j of the first principal component and the integrated determination value P2 _j of the second principal component is the maximum is the human action / behavior at the measurement time. May be identified.

  The identification method, identification apparatus, and identification program for human behavior of the present invention can accurately identify the behavior of a person in a facility. For example, residents (especially elderly people who are living alone or physically disabled people) High utility value in fields such as daily watching and support, and rescue in the event of a natural disaster.

10 Human Behavior Identification Device 17 Human Behavior Identification Program

Claims (6)

A frequency distribution of measurement values of reference data acquired as a time history waveform by measurement related to operation information for each behavior mode is created, and the reference data is subjected to Fourier transform and the frequency of spectrum values for each frequency component after the Fourier transform. A distribution is created, and a frequency distribution of measurement values of observation data acquired as a time history waveform by measurement related to the motion information in the action identification target time zone is created, and the observation data is Fourier transformed and the Fourier transform concerned A frequency distribution of spectrum values for each subsequent frequency component is created, and accordingly, a combination data group of a class attribute of the measurement value and a frequency value for each action mode related to the reference data, the frequency component and the spectrum Value combination data group, class attribute of attribute value and frequency value A combination data group, a combination data group of the measurement value class attribute and the frequency value regarding the observation data, a combination data group of the frequency component and the spectrum value, and a class attribute of the spectrum value A combination data group with frequency values is prepared, and a first determination index calculated by Equation 1 and Equation 2 for each of the three types of combination data groups for each of the reference data and the observation data The at least two determination index values of the value, the second determination index value calculated by Formula 3 and Formula 4, and the third determination index value calculated by Formula 5 are used to calculate Formula 6 to Formula 8. An integrated judgment value is calculated for each behavior mode, and the behavior identification target time zone is calculated based on the integrated judgment value for each behavior mode. Method of identifying a human behavior, characterized in that the type of behavior of the person is identified.
Where Χ: evaluation value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
Where P: first determination index value,
χ ₁ ² (Χ _{d, j} ): represents a chi-square distribution value with one degree of freedom corresponding to Χ _{d, j} .
Where, B: second determination index value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
However, Equation 4 holds for N (Ba _{d, j, k} ) and N (Ob _{d, k} ).
Where R: third determination index value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
Moreover, Ba _{d, j} ¯ (correctly, Ba _d, on the _j - stick) is an average value for all of the k of _{Ba d, j, k, Ob} d ¯ ( correctly, on the Ob _d (-) Is an average value for all _k of Ob _{d, k} .
Where Pr: integrated judgment value,
I: first determination index value, second determination index value, or third determination index value,
C: weighting factor,
m: identifier indicating the type of determination index value,
j: identifier indicating the type of action mode,
Ni: Represents the number of types of determination index values used.
Further, I _m ￣ (correctly, “-” is added on I _m ) is an average value for all _j of I _{j, m} .
However, C is determined so as to maximize Sp ² represented by Expression 8 using Expression 7 as a constraint.
Where Sp ^{2 is the} amount of dispersion,
no: represents the number of action modes.
In addition, Pr 正 し く (correctly, “-” is attached on Pr) is an average value for all _j of Pr _j .   2. The vibration displacement data, vibration speed data, vibration acceleration data, or strain data acquired as a time history waveform is used as the reference data and the observation data. A method for identifying the described human behavior. Means for creating a frequency distribution of measured values of reference data acquired as a time history waveform by measurement relating to operation information for each behavior mode; means for Fourier transforming the reference data; and frequency components after Fourier transform of the reference data Means for creating a frequency distribution of each spectrum value, means for creating a frequency distribution of measurement values of observation data acquired as a time history waveform by measurement relating to the motion information in a behavior identification target time zone, and the observation data Means for performing a Fourier transform, and means for creating a frequency distribution of spectral values for each frequency component after Fourier transform of the observation data, and by these means, the class of the measured values for each action mode related to the reference data A combination data group of attribute values and frequency values, and combination data of the frequency component and the spectrum value Group, a combination data group of the spectrum value class attribute and the frequency value, and a combination data group of the measurement value class attribute and the frequency value regarding the observation data, the frequency component and the spectrum value And a combination data group of the class attribute of the spectrum value and the frequency value, and further, the three types of the reference data and the observation data according to Equation 1 and Equation 2 A means for calculating a first determination index value for each type of combination data group, and a second for each type of the three types of combination data groups related to each of the reference data and the observation data according to Equation 3 and Equation 4. Means for calculating a determination index value, and the three types relating to each of the reference data and the observation data according to Equation 5 Means for calculating a third determination index value for each type of combination data group; and the first to third determination indexes for each type of the three types of combination data groups with respect to each of the reference data and the observation data Means for calculating an integrated determination value for each behavior mode using Equations 6 to 8 using at least two determination index values of the values; and the action identification target time zone based on the integrated determination value for each behavior mode And a means for identifying the type of human behavior in the human behavior identification device.
Where Χ: evaluation value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
Where P: first determination index value,
χ ₁ ² (Χ _{d, j} ): represents a chi-square distribution value with one degree of freedom corresponding to Χ _{d, j} .
Where, B: second determination index value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
However, Equation 4 holds for N (Ba _{d, j, k} ) and N (Ob _{d, k} ).
Where R: third determination index value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
Moreover, Ba _{d, j} ¯ (correctly, Ba _d, on the _j - stick) is an average value for all of the k of _{Ba d, j, k, Ob} d ¯ ( correctly, on the Ob _d (-) Is an average value for all _k of Ob _{d, k} .
Where Pr: integrated judgment value,
I: first determination index value, second determination index value, or third determination index value,
C: weighting factor,
m: identifier indicating the type of determination index value,
j: identifier indicating the type of action mode,
Ni: Represents the number of types of determination index values used.
Further, I _m ￣ (correctly, “-” is added on I _m ) is an average value for all _j of I _{j, m} .
However, C is determined so as to maximize Sp ² represented by Expression 8 using Expression 7 as a constraint.
Where Sp ^{2 is the} amount of dispersion,
no: represents the number of action modes.
In addition, Pr 正 し く (correctly, “-” is attached on Pr) is an average value for all _j of Pr _j .   4. The vibration displacement data, vibration speed data, vibration acceleration data, or strain data acquired as a time history waveform is used as the reference data and the observation data. The identification device for the human behavior described. Processing for creating a frequency distribution of measurement values of reference data acquired as a time history waveform by measurement related to operation information for each behavior mode, processing for Fourier transforming the reference data, and frequency components after Fourier transform of the reference data A process of creating a frequency distribution of each spectrum value, a process of creating a frequency distribution of measurement values of observation data acquired as a time history waveform by measurement related to the motion information in the action identification target time zone, and the observation data A process for performing a Fourier transform and a process for creating a frequency distribution of spectral values for each frequency component after Fourier transform of the observation data are performed by a computer, and by these processes, the measurement values for the action modes related to the reference data A combination data group of class attributes and frequency values, the frequency component and the spectrum value A combination data group, a combination data group of the spectrum value class attribute and the frequency value, and a combination data group of the measurement value class attribute and the frequency value regarding the observation data, the frequency component and the frequency data A combination data group of spectrum values and a combination data group of class attribute and frequency value of the spectrum value are prepared, and further, the three data relating to each of the reference data and the observation data are expressed by Expression 1 and Expression 2, respectively. A process for calculating a first determination index value for each type of the combination data group of types, and a third for each type of the three types of combination data groups for each of the reference data and the observation data according to Equation 3 and Equation 4. A process for calculating a second determination index value and each of the reference data and the observation data by Equation 5 A process for calculating a third determination index value for each type of the three types of combination data groups, and the first to third types for each of the three types of combination data groups for the reference data and the observation data. Based on the integrated determination value for each behavior mode, the processing for calculating the integrated determination value for each behavior mode by Equations 6 to 8 using at least two determination index values of the third determination index value A human behavior identification program which causes a computer to perform processing for identifying a type of human behavior in a behavior identification target time zone.
Where Χ: evaluation value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
Where P: first determination index value,
χ ₁ ² (Χ _{d, j} ): represents a chi-square distribution value with one degree of freedom corresponding to Χ _{d, j} .
Where, B: second determination index value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
However, Equation 4 holds for N (Ba _{d, j, k} ) and N (Ob _{d, k} ).
Where R: third determination index value,
Ba: Frequency value or spectrum value related to the reference data,
Ob: Frequency value or spectrum value for observation data,
d: identifier indicating the type of combination data group,
j: identifier indicating the type of action mode,
k: identifier indicating class attribute type or frequency component type,
n (d): Type of combination data group (d) Total number of class attributes by class or
Represents the total number of frequency components.
Moreover, Ba _{d, j} ¯ (correctly, Ba _d, on the _j - stick) is an average value for all of the k of _{Ba d, j, k, Ob} d ¯ ( correctly, on the Ob _d (-) Is an average value for all _k of Ob _{d, k} .
Where Pr: integrated judgment value,
I: first determination index value, second determination index value, or third determination index value,
C: weighting factor,
m: identifier indicating the type of determination index value,
j: identifier indicating the type of action mode,
Ni: Represents the number of types of determination index values used.
Further, I _m ￣ (correctly, “-” is added on I _m ) is an average value for all _j of I _{j, m} .
However, C is determined so as to maximize Sp ² represented by Expression 8 using Expression 7 as a constraint.
Where Sp ^{2 is the} amount of dispersion,
no: represents the number of action modes.
In addition, Pr 正 し く (correctly, “-” is attached on Pr) is an average value for all _j of Pr _j .   6. The vibration displacement data, vibration speed data, vibration acceleration data, or strain data acquired as a time history waveform is used as the reference data and the observation data. Identification program for human behavior described.