JP2009134590A - Action identification system, action identification method, optimum sensor set determination method and optimum parameter determination method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an action identification system capable of accurately identifying complicated actions. <P>SOLUTION: The action identification system 10 includes a server 12, and the server 12 identifies the action of a subject. When receiving detection data from a passing sensor 80, the server 12 specifies the present position (location) of the subject on the basis of the detection data. Then, the server 12 reads the optimum combination of movement measuring apparatuses 60 corresponding to the specified location and an optimum parameter set to each of the plurality of movement measuring apparatuses 60 included in it from an optimum sensor parameter DB 18 and transmits them to a repeater 14 carried by the subject. Then, the repeater 14 stops the transmission of acceleration data by the apparatuses other than the optimum movement measuring apparatus 60 and sets the optimum parameter to the acceleration sensor (64) of the optimum movement measuring apparatus 60. The server 12 receives the acceleration data detected by the optimum parameter from the optimum movement measuring apparatus 60 from the repeater 14 and identifies one action from an action group which can be taken in the location as the action of the subject. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a behavior identification system, a behavior identification method, an optimal sensor set determination method, and an optimal parameter determination method, and in particular, for example, a behavior identification system that identifies one behavior that a subject can perform from a plurality of behavior groups that differ from place to place. The present invention relates to a behavior identification method, an optimal sensor set determination method, and an optimal parameter determination method.

  An example of a conventional action identification system of this type is disclosed in Patent Document 1. In the cellular phone device disclosed in Patent Document 1, when the owner designates a desired action (for example, “walking 1”) on the registration screen and inputs a start using the key input unit while performing the action, the mobile phone apparatus is designated in advance. The acceleration sensor unit detects the motion acceleration for the time, and the reference behavior pattern of the behavior is stored in the pattern storage unit. When the operation mode is activated after such registration, the acceleration sensor unit detects motion acceleration due to the owner's behavior, and the behavior detection unit compares it with the reference behavior pattern to identify the current behavior pattern of the subject. To do.

Another example of this type of behavior identification system is disclosed in Patent Document 2. In the motion information measurement system of Patent Document 2, motion data is collected from a motion information measurement device mounted on the user's wrist, upper arm, knee, waist, both ankles, and the like. This motion information-related measuring device is composed of an acceleration sensor and / or a gyro sensor, and measures acceleration and / or angular velocity due to a user's motion at the wearing site, or both. The recognition of the action is performed based on the action recognition rule that the action information comparison and recognition unit has. For example, when the information from the motion information measuring device attached to the waist has a small amount of exercise and the information from the motion information measuring device attached to the wrist has a large amount of exercise, it recognizes that the motion is in a stationary state. Conversely, when the amount of exercise from the motion information measuring device attached to the waist is large, it is recognized that the exercise is a whole body exercise.
JP2003-46630 [H04M 1/247, H04Q 7/38] JP 2004-184351 [G01B 21/00, A61B 5/00, A61B 5/0488, A61B 5/11, G06F 3/00]

  However, in the technique of Patent Document 1, “walking”, “sitting”, “standing”, “boarding a vehicle” are performed according to the time transition of acceleration in the X, Y, and Z directions of the acceleration sensor unit provided in the mobile phone device. However, it is difficult to recognize complex behavior because the behavior is identified using only the acceleration from one acceleration sensor. In the technique of Patent Document 2, the momentum of each part is detected from the action data such as acceleration from the action information measuring device attached to a plurality of parts, and the whole body exercise or the movement in a stationary state is performed. Although the actions are roughly classified as to whether or not there is, the actions are identified using only the action data such as the acceleration, so that it is difficult to recognize the complicated actions.

  Therefore, a main object of the present invention is to provide a novel behavior identification system, behavior identification method, optimal sensor set determination method, and optimal parameter determination method.

  Another object of the present invention is to provide a behavior identification system, a behavior identification method, an optimal sensor set determination method, and an optimal parameter determination method that can accurately identify a complex behavior of a subject.

  The present invention employs the following configuration in order to solve the above problems. Note that reference numerals in parentheses, supplementary explanations, and the like indicate correspondence relationships with embodiments described later to help understanding of the present invention, and do not limit the present invention in any way.

  1st invention is an action identification system which identifies one action which a subject can perform from a plurality of action groups which differ according to a place, and is the optimal among a plurality of motion measuring devices with which a subject's different part was equipped. Storage means for storing a combination of various motion measurement devices in association with each location, position detection means for detecting the current position of the subject, and a motion measurement device corresponding to a location including the current location of the subject detected by the position detection means A reading means for reading the combination from the storage means as an optimum sensor set, a movement detection means for detecting outputs from a plurality of movement measuring devices included in the optimum sensor set read by the reading means, and a subject from the detection result of the movement detection means It is an action identification system provided with the action identification means which identifies these actions.

  In the first invention, the behavior identification system (10) identifies the behavior of the subject based on the current position of the subject and the outputs from the plurality of motion measurement devices (60) attached to different parts of the subject. The behavior identification system includes a storage unit (18), and the storage unit stores an optimal combination of motion measurement devices selected from a plurality of motion measurement devices in association with each location. The position detecting means (12, 80, S51) detects the current position of the subject, and the reading means (12, 18, S53) uses the combination of the motion measuring devices corresponding to the place including the current position as the optimum sensor set. Read from storage means. The motion detection means (40, 60, S87) detects outputs from a plurality of motion measurement devices included in the optimum sensor set. In the embodiment, the acceleration corresponding to the movement of each part of the subject is acquired by the acceleration sensor (64) included in the motion measuring device. And an identification means (12, S61) identifies a test subject's action using the output from the several motion measurement apparatus contained in the optimal sensor set.

  According to the first invention, the action is identified based on the acceleration data obtained from the motion measuring device included in the optimum sensor set determined according to the place where the subject exists. Therefore, the complex behavior of the subject can be identified more accurately.

  A second invention is an action identification system that identifies one action that a subject can perform from a plurality of action groups that differ depending on a location, and is set in each of a plurality of motion measurement devices attached to different parts of the subject Storage means for storing the optimum parameters to be associated with each location, position detection means for detecting the current position of the subject, and storage means for storing the optimal parameters corresponding to the location including the current position of the subject detected by the position detection means Reading means for reading from, parameter setting means for setting the optimum parameters read by the reading means in each of the plurality of motion measuring devices, and motion for detecting outputs from the plurality of motion measuring devices set with the parameters by the parameter setting means A behavior identification system comprising behavior identification means for identifying the behavior of the subject from the detection results of the detection means and the motion detection means; Is Temu.

  In the second invention, the behavior identification system (10) identifies the behavior of the subject based on the current position of the subject and the outputs from the plurality of motion measurement devices (60) attached to different parts of the subject. The behavior identification system includes a storage unit (18), and the storage unit stores optimum parameters set in each of the plurality of motion measurement devices in association with each location. In an embodiment, the parameter is a sampling frequency at which the motion measurement device acquires sensor data. The position detection means (12, 80, S51) detects the current position of the subject, and the reading means (12, 18, S53) reads out the parameter corresponding to the location including the current position from the storage means, for example, as the optimum parameter. . The behavior identification system includes parameter setting means (40, 60, S85) and motion detection means (40, 60, S87). The parameter setting means sets the parameters of the plurality of motion measuring devices to values indicated by the optimum parameters read by the reading means. The motion detection means (40, 60, S87) detects outputs from a plurality of motion measurement devices. In the embodiment, the acceleration corresponding to the movement of each part of the subject is acquired by the acceleration sensor (64) included in the motion measuring device. The motion detection means detects sensor data output from the motion measurement device. And an identification means (12, S61) identifies a test subject's action using the output from the several exercise | movement measuring device set to the optimal parameter.

  According to the second invention, the action is identified based on the acceleration data obtained from the motion measurement device in which the optimum parameter determined for each place including the current position of the subject is set. Therefore, the complex behavior of the subject can be identified more accurately.

  3rd invention is an action identification system which identifies one action which a subject can perform from a plurality of action groups which differ according to a place, and is the optimal among a plurality of movement measuring devices with which a subject's different part was equipped. First storage means for storing a combination of various motion measurement devices in association with each location, and associating the optimum parameters set in each of the plurality of motion measurement devices included in the optimal motion measurement device combination with each location A combination of a second storage unit that stores the current position of the subject, a position detection unit that detects the current position of the subject, and a motion measurement device that corresponds to the location that includes the current position of the subject detected by the position detection unit is stored as an optimal sensor set. First reading means for reading out from the means, and the optimum set for each of the plurality of motion measuring devices included in the optimum sensor set read out by the first reading means Included in the optimum sensor set, the second readout means for reading out the parameters from the second storage means, the parameter setting means for setting the optimum parameters read out by the second readout means in each of the plurality of motion measuring devices included in the optimum sensor set It is an action identification system provided with the action detection means which detects the output from the several exercise | movement measuring device to be detected, and the action identification means which identifies a test subject's action from the detection result of a movement detection means.

  In the third invention, the behavior identification system (10) identifies the behavior of the subject based on the current position of the subject and the outputs from the plurality of motion measurement devices (60) attached to different parts of the subject. The action identification system includes a first storage means (18) and a second storage means (18). In the first storage means, an optimum combination of motion measurement devices selected from a plurality of motion measurement devices is stored for each location. In the second storage means, the optimum parameters set in each of the plurality of motion measuring devices are stored in association with each location. In an embodiment, the parameter is a sampling frequency at which the motion measurement device acquires sensor data. The position detection means (12, 80, S51) detects the current position of the subject. The first reading means (12, 18, S53) reads the combination of the motion measurement devices corresponding to the place including the current position from the first storage means as the optimum sensor set, and the second reading means (12, 18, S53). ) Reads out the parameters set in each of the plurality of motion measuring devices included in the optimum sensor set read by the first reading means from the second storage means as the optimum parameters. The behavior identification system includes parameter setting means (40, 60, S85) and motion detection means (40, 60, S87). The parameter setting means sets the parameters of the plurality of motion measurement devices included in the optimum sensor set to values indicated by the optimum parameters read by the second reading means. The motion detection means (40, 60, S87) detects outputs from a plurality of motion measurement devices included in the optimum sensor set. In the embodiment, the acceleration corresponding to the movement of each part of the subject is acquired by the acceleration sensor (64) included in the motion measuring device. And an identification means (12, S61) identifies a test subject's action using the output from the several motion measurement apparatus contained in the optimal sensor set.

  According to the third aspect, the optimum parameter determined for each place including the current position of the subject is set, and the action is identified based on the acceleration data obtained from the motion measurement device included in the optimum sensor set. Therefore, the complex behavior of the subject can be identified more accurately.

  4th invention is dependent on 1st or 3rd invention, and is further provided with the pause means which pauses the motion measurement apparatus which is not contained in the optimal sensor set.

  In 4th invention, the action identification system (10) is further provided with a pause means (40, 60, S83). In the pause means, for example, with reference to the optimum sensor set from the storage means (18), the motion measurement device (60) not included in the optimum sensor set in the place including the current position of the subject is paused. In the embodiment, the motion measurement device is switched to the sniff mode, and the transmission of the acceleration data is paused.

  According to the fourth aspect of the invention, it is possible to reduce the power consumption of the motion measurement device that is not included in the optimum sensor set.

  A fifth invention is dependent on any one of the first to fourth inventions, and a class classifier is provided by associating a reference behavior feature amount acquired when a subject performs a plurality of actions in advance with each action. Learning means for learning is further provided, and the behavior identification means uses the extraction means for extracting the subject's behavior feature amount from the detection result of the motion detection means, and the behavior extracted by the extraction means using the class classifier learned by the learning means. Action specifying means for specifying the action of the subject from the feature amount is included.

  In the fifth invention, the behavior identification system (10) further includes a learning means (12), and the learning means assigns, to each action, a reference action feature amount acquired when the subject executes a plurality of actions in advance. Correlate and learn the classifier. In this embodiment, SVM (Support Vector Machine) is adopted as a classifier. The action identifying means (12, S59, S61) includes an extracting means (12, S59) and an action specifying means (12, S61). The extraction means extracts the behavior feature amount of the subject from the detection result of the motion detection means (40, 60, S87). In the embodiment, the time-series acceleration data is divided into a sliding window of a fixed time, and five types of feature quantities of average, standard deviation, energy, frequency domain entropy, and correlation coefficient are obtained for each window. Then, the behavior identifying means identifies the behavior of the subject from the behavior feature amount extracted by the extracting means, using the class classifier learned by the learning means.

  6th invention is equipped with the memory | storage means which matches and memorize | stores the optimal combination of an exercise | movement measurement apparatus for every place among the several exercise | movement measurement apparatuses with which the test subject was mounted | worn, and the computer which identifies a test subject's action (A) detecting the current position of the subject, and (b) selecting the optimal combination of motion measuring devices corresponding to the location including the current position of the subject detected in step (a). Read out from the storage means as a sensor set, (c) detect outputs from a plurality of motion measuring devices included in the optimal sensor set read out in step (b), and (d) detect the test subject from the detection result in step (c) This is an action identification method for identifying an action.

  The sixth invention also has the same effect as the first invention.

  7th invention is equipped with the memory | storage means which matches and memorize | stores the optimal parameter set to each of several exercise | movement measuring apparatus with which the test subject's different site | part was mounted | worn for every place, The action of a computer which identifies test subject's action (A) a current position of the subject is detected, (b) an optimum parameter corresponding to the location including the current position of the subject detected in step (a) is read from the storage means, and (c) step The optimum parameters read in (b) are set in each of the plurality of motion measurement devices, (d) the outputs from the plurality of motion measurement devices are detected, and (e) the subject's behavior is detected from the detection result in step (d). It is an action identification method for identifying

  The seventh invention also has the same effect as the second invention.

  According to an eighth aspect of the present invention, there is provided a first storage unit for storing a combination of optimal motion measurement devices in association with each location among a plurality of motion measurement devices mounted on different parts of the subject, and an optimal motion measurement device. A second storage means for storing an optimum parameter set in each of a plurality of motion measurement devices included in the combination in association with each place, and a computer behavior identification method for identifying a subject's behavior, a) Detecting the current position of the subject, (b) Reading the optimum combination of motion measurement devices corresponding to the location including the current position of the subject detected in step (a) from the first storage means as the optimum sensor set (C) Read the optimum parameters in the plurality of motion measurement devices included in the optimum sensor set read in step (b) from the second storage means as the optimum parameters, and (d) step The optimum parameters read in step (c) are set in each of the plurality of motion measurement devices included in the optimum sensor set, and (e) a plurality of motion measurements included in the optimum sensor set in which the parameters are set in step (d). This is an action identification method for detecting an output from the apparatus and (f) identifying the action of the subject from the detection result of step (e).

  The eighth invention also has the same effect as the third invention.

  A ninth invention is an optimal sensor set for determining an optimal combination of motion measurement devices associated with each location from a plurality of motion measurement devices attached to different parts of the subject in order to identify the subject's behavior A determination method, wherein (a) a first feature value is extracted from motion data corresponding to a group of actions performed by a subject at a certain location, and (b) a main feature about the first feature value extracted in step (a) Based on the component analysis, a second feature value extracted from the first feature value with an element having a contribution ratio larger than a predetermined threshold is selected, and (c) the second feature value selected in step (b) is included. (D) Steps (a) to (c) are repeated until there are no more exclusion targets in step (c), and (e) a combination of motion measurement devices that have not been excluded. As the optimal sensor set This is an optimal sensor set determination method.

  In 9th invention, in order to identify a test subject's action, the combination of the optimal motion measurement apparatus matched for every place is determined from the several motion measurement apparatus (60) with which the said test subject was mounted | worn with the different site | part. In step (a), a first feature value is extracted from motion data corresponding to a group of actions performed by a subject at a certain place. However, the motion data in the embodiment is acceleration data corresponding to the movement of each part of the subject acquired by the acceleration sensor (64) included in the motion measuring device. In step (b), the extracted first feature quantity is subjected to principal component analysis, and a second feature quantity in which an element having a contribution rate larger than a predetermined threshold is extracted from the first feature quantity is selected. In the embodiment, the threshold is set to 1%. In step (c), the motion measurement device that does not have the selected second feature is removed from the set of motion measurement devices to be used at that location. However, at this time, if there is no motion measurement device that does not have the selected feature amount, all the motion measurement devices become the optimal sensor set at the location. In step (d), feature vectors are extracted from the acceleration data of the set of motion measurement devices excluding the removed motion measurement device, and a principal component analysis is performed to select a feature amount having a large contribution rate. Then, the motion measurement device that does not have the selected feature amount is removed from the set of motion measurement devices used at that location. Such a process is repeated until there is no motion measurement device that does not have the selected feature amount. In step (e), a combination of motion measurement devices that have not been removed is determined as an optimal sensor set at the location.

  According to the ninth aspect, an optimum sensor parameter database can be constructed.

  In a tenth aspect of the present invention, an optimum parameter associated with each location is set in a motion detection sensor provided in each of a plurality of motion measurement devices attached to different parts of the subject in order to identify the subject's behavior. An optimal parameter determination method for determining, comprising: (a) a plurality of motion data detected by a plurality of motion detection sensors set at a predetermined sampling frequency and corresponding to a group of actions performed by a subject at a certain location; Resample by changing the sampling frequency of each motion detection sensor, (b) extract feature values from the resampled motion data in step (a), and (c) identify feature values extracted in step (b) Errors are calculated, and (d) steps (a) through steps are calculated until identification errors are calculated for all sampling frequencies within a predetermined range. (C) repeated, and the identification error calculated in step (e) (c) the minimum, and determines the combinations of the sampling frequency as an optimal parameter, the optimal parameter determining method.

  In the tenth invention, in order to identify the behavior of the subject, the optimum associated with each location set in the motion detection sensor provided in each of the plurality of motion measurement devices (60) attached to different parts of the subject The correct parameters. However, the motion detection sensor in the embodiment is an acceleration sensor (64) included in the motion measurement device. In step (a), a plurality of motion data corresponding to a group of actions performed by a subject at a certain location, detected by a plurality of motion detection sensors used at a certain location, set to a predetermined sampling frequency. Resampling is performed by changing the sampling frequency of each motion detection sensor. In the embodiment, the sampling frequency set in the acceleration sensor is changed stepwise from 50 Hz to 25 Hz, 12.5 Hz, 6.25 Hz, and 3.125 Hz, and the acceleration data is resampled. In step (b), feature quantities are extracted from the sampled motion data. In step (d), an identification error is calculated for the extracted feature quantity. In step (e), identification errors are calculated for all sampling frequencies within a predetermined range. In step (f), the sampling frequency that minimizes the calculated identification error is determined as the optimum parameter at the location.

  According to the tenth aspect, an optimum sensor parameter database can be constructed.

  According to the present invention, the behavior is identified based on the outputs of the plurality of motion measurement devices included in the optimum sensor set determined according to the location where the subject exists, and thus the complex behavior of the subject is more accurately identified. be able to.

  The above object, other objects, features and advantages of the present invention will become more apparent from the following detailed description of embodiments with reference to the drawings.

  Referring to FIG. 1, a behavior identification system 10 according to an embodiment of the present invention includes a server 12 that also functions as a behavior identification device. The behavior identification system 10 is applied to a certain environment and performs behavior of a person (subject) in the environment. Identify. In this embodiment, the behavior identification system 10 is provided in a hospital building and identifies the behavior of a nurse as a subject, but is not limited thereto.

  The server 12 is connected to the repeater 40 and a plurality of passage sensors 80 that the subject wears or possesses via a wired or wireless communication line (hereinafter, network 30). In addition, a plurality of motion measurement devices 60 attached to the subject are communicably connected to the repeater 40. In FIG. 1, one repeater 40 is shown for simplicity. However, since there are a plurality of nurses wearing the repeater 40, for example, the same number of repeaters 40 as the plurality of nurses. Is also provided. A plurality of motion measuring devices 60 are connected to each repeater 40.

  A plurality of databases (hereinafter referred to as DB) are connected to the server 12. Specifically, the server 12 is connected to a sensor DB 14, a knowledge DB 16, an optimum sensor parameter DB 18, and an action record DB 20.

  As will be described later, in the sensor DB 14, acceleration data from the motion measurement device 60 corresponding to the behavior of the subject and data on the position of the subject specified according to the detection data from the passage sensor 80 (position data) are respectively in time series. Remembered.

  The knowledge DB 16 stores in advance data (label data) about action names and action identification information (behavior labels) included in each action group for action groups (behavior classes) that the subject can take according to the location. It is. However, since the procedure and habit of behavior may differ for each subject, label data may be stored for each subject.

  The optimum sensor parameter DB 18 stores an optimum sensor set and optimum parameters associated with each location. Here, the optimal sensor set means a combination of the motion measurement devices 60 used to identify one action from a group of actions that can be taken by the subject. Specifically, the identification information relating to the combination of the plurality of motion measurement devices 60 and the names of the parts to which the plurality of motion measurement devices are attached at each location are stored in the optimum sensor parameter DB 18 corresponding to each location. Is done. The optimum parameter means a parameter set for each of the plurality of motion measuring devices 60 to be used. Specifically, similarly to the optimum sensor set, parameters set in each of the plurality of motion measurement devices 60 used in the corresponding place are stored in the optimum sensor parameter DB 18 corresponding to each place. However, since only the parameters set for each of the plurality of motion measuring devices 60 used are stored as the optimum parameters, the parameters set for each of the plurality of motion measuring devices 60 used are stored. Can be said to be remembered.

  As will be described in detail later, the optimal sensor parameter DB 18 is constructed by learning using the position (place) of the subject and the motion data of the subject acquired in advance when the subject acts ( (See FIGS. 9, 10 and 11).

  In the behavior record DB 20, as will be described in detail later, the behavior of the subject identified by the server 12 is stored, for example, in time series.

  In this embodiment, the repeater 40 is mounted on the waist of the subject (see FIG. 3). The repeater 40 is attached to the subject with a fixture such as a rubber band. However, the repeater 40 may be carried by the subject in a bag, waist pouch or rucksack used by the subject. In this embodiment, since the subject may move, the network 30 is provided with a plurality of access points (not shown), and the repeater 40 is connected to the network 30 via any one of the access points. The

  FIG. 2 is a block diagram showing a specific configuration of the repeater 40, and the repeater 40 includes a CPU 42. An interface 44, a RAM 46 and a Bluetooth (registered trademark) module 48 are connected to the CPU 42. An antenna 50 is connected to the Bluetooth module 48. In the repeater 40, acceleration data transmitted from the motion measuring device 60 is provided to the CPU 42 via the antenna 50 and the Bluetooth module 40.

  The CPU 42 stores (temporarily stores) acceleration data from the motion measuring device 60 in the RAM 46, and at a timing when connected to the network 30, the CPU 42 converts the acceleration data stored in the RAM 46 into an interface 44 such as a LAN (wireless LAN) adapter. And transmitted to the server 12 via the network 30. The server 12 receives acceleration data from the network 30 via an interface (not shown) such as a LAN adapter, and stores (registers) the received acceleration data in the sensor DB 14.

  The CPU 42 of the repeater 40 receives the optimum sensor set identification information and the optimum parameters from the server 12. Specifically, the server 12 reads the optimum sensor set and optimum parameters corresponding to the location including the position of the repeater 40 (the subject who owns it) from the optimum sensor parameter DB 18 and transmits the optimum sensor set and the optimum parameters to the repeater 40. On the other hand, when the repeater 40 receives the optimum sensor set and the optimum parameters transmitted from the server 12, the repeater 40 causes the motion measurement device 60 not included in the optimum sensor set to stop transmitting the acceleration data according to the optimum sensor set and the optimum parameter. The acceleration sensor 64 (see FIG. 4) of the motion measuring device 60 included in the set is set to the sampling frequency indicated by the optimum parameter.

  As described above, the repeater 40 is communicably connected to the plurality of motion measurement devices 60 (60a, 60b, 60c, 60d) attached to the subject's body. Specifically, as shown in FIG. 3, the motion measurement device 60a is attached to the right arm, the motion measurement device 60b is attached to the left arm, the motion measurement device 60c is attached to the chest, and the motion measurement device 60d is attached to the waist. Is done. Although illustration is omitted, each motion measuring device 60 is attached to the subject with a fixing tool such as a rubber band. Although not shown, in this embodiment, the motion measuring device 60 is driven by a battery (not shown).

  FIG. 4 is a block diagram showing a specific configuration of the motion measuring device 60, and the motion measuring device 60 includes a CPU 62. An acceleration sensor 64, a RAM 66, and a Bluetooth module 68 are connected to the CPU 62. An antenna 70 is connected to the Bluetooth module 68. The acceleration sensor 64 is, for example, a multi-axis (three-axis) acceleration sensor, and can be variably set within a sampling frequency of 0 to 200 Hz, and the acceleration can be measured up to ± 3 G (G is gravity).

  In the motion measuring device 60, acceleration data output from the acceleration sensor 64 is given to the CPU 62. The CPU 62 stores (temporarily stores) the acceleration data from the acceleration sensor 64 in the RAM 66, and the acceleration data for the certain time is transmitted via the Bluetooth module 68 and the antenna 70 every certain time (for example, 10 seconds). It transmits to the repeater 40. Although detailed description is omitted, it is determined in advance which part of the subject each motion measurement device 60 is to be worn, and the acceleration data transmitted from each motion measurement device 60 includes the motion measurement. Identification information (ID) for identifying the device 60 is attached as a label. For example, the MAC address of the Bluetooth module 68 can be used as the ID. However, a ROM that can be accessed by the CPU 62 may be provided inside the motion measuring device 60, and an ID may be stored in the ROM.

  On the other hand, although not described in the above description, the knowledge DB 16 stores a table in which the part to which the motion measuring device 60 is to be attached is described corresponding to the ID of the motion measuring device 60. Therefore, the server 12 that has received the acceleration data transmitted from the repeater 40 refers to the knowledge DB 16 from the label (ID) attached to the acceleration data from the motion measurement device 60 attached to any part. Whether the data is acceleration data is identified, and the acceleration data is stored in the sensor DB 14 in time series.

  As described above, the CPU 62 of the motion measuring apparatus 60 receives a command for instructing to pause transmission of acceleration data and a command for setting optimum parameters from the repeater 40 via the antenna 70 and the Bluetooth module 68. Given. The CPU 62 controls the detection of acceleration data in accordance with the command from the repeater 40.

  The passage sensor 80 detects an infrared signal (transmitter ID) from the transmitter 100 attached to the subject and notifies the server 12 of the infrared signal. As shown in FIG. 3, the transmitter 100 is attached to the head (hair) of a subject using a fixing device such as a rubber band. The reason why the transmitter 100 is mounted on the user's head is that, as will be described later, a light receiving module (84, 86) that receives (receives) the transmitter ID from the transmitter 100 is provided above the gate. (See FIG. 6). Therefore, depending on the mounting position of the light receiving module (84, 86), it is possible to attach it to other parts such as the waist, wrist or arm of the subject.

  FIG. 5 is a block diagram showing a specific configuration of the passage sensor 80, and the passage sensor 80 includes a CPU 82. The CPU 82 is connected to an interface 90 such as a first light receiving module 84, a second light receiving module 86, a RAM 88 and a LAN (wireless LAN) adapter. As shown in FIG. 6, the first light receiving module 84 and the second light receiving module 86 are the upper part of the gate and are mounted on different side surfaces. Further, in the case 92 placed on the gate, components (82, 88, 90) other than the light receiving modules (84, 86) of the passage sensor 80 are incorporated. However, the gate is only shown for easy understanding. For example, as shown in FIG. 7, a nurse station, an examination room, a doorway of a room such as a hospital room, a connecting portion between a hallway and a hallway (crossroad, T-shaped The passage sensor 80 is installed at a key point such as a connecting portion between a corridor and a staircase.

  Although detailed explanation is omitted, unique identification information (light receiving module ID) is assigned to each of the light receiving modules (84, 86) of each passage sensor 80. Each light receiving module (84, 86) detects an infrared signal from the transmitter 100. Here, the infrared signal transmitted from the transmitter 100 is, for example, an 8-bit fixed pattern (transmitter ID). For example, when receiving the transmitter ID, the light receiving module (84, 86) gives the received transmitter ID to the CPU 82. The CPU 82 adds the light receiving module ID of the light receiving module (84, 86) that has received the transmitter ID, and stores (temporarily stores) it in the RAM 88, and then the server 82 via the interface 90 such as a LAN adapter and the network 30. 12 to send.

  On the other hand, when the server 12 receives the transmitter ID to which the light receiving module ID transmitted from the passage sensor 80 is added, the server 12 refers to the knowledge DB 16 and determines the subject and the presence from the received light receiving module ID and the transmitter ID. Specify the location (current location).

  The server 12 can specify not only the current position of the subject but also its movement. For example, when a specific transmitter ID to which the light receiving module ID of the first light receiving module 84 is added is detected and then the transmitter ID to which the light receiving module ID of the second light receiving module 86 is added is detected. , It can be determined that the subject indicated by the transmitter ID has moved from its current position.

  By the way, although not shown, each nurse performs each nursing operation according to a schedule (worksheet) for the daily operation. For example, FIG. 8 is performed by a nurse by monitoring the behavior of a plurality of nurses on a floor (see FIG. 7) of a hospital ward to which the behavior identification system 10 of this embodiment is applied. This is an occurrence distribution of work actions (contents of nursing work) at work places (places where nursing work is performed) obtained by detecting each nursing work for each place. However, although omitted in FIGS. 1 to 3, the subject is provided with a pin microphone with a button switch, and voice data corresponding to the subject's voice input through the pin microphone is given to the repeater 40. Then, the nurse inputs the contents of the work (nursing work) and the start and end thereof through the pin microphone. Therefore, identification information (action label) can be given to each action recorded by using the action identification system 10 as described above based on the input voice. However, regardless of the voice, the administrator of the server 12 refers to the nurse's worksheet and daily report, or refers to the nurse's action taken with the video camera. A behavioral label may be added.

  In FIG. 8, the sum of the occurrence probability of each work action is 1 on the vertical axis (location). Each plot is shown in a color closer to black as the occurrence probability is higher. As shown in FIG. 8, for example, when the work place is “ophthalmic examination room”, “walking / moving”, “document work”, “ophthalmic examination assistance”, “ophthalmic examination assistance preparation”, “hygiene processing” “Patient transfer” is a group of actions that can be taken by the subject, and other actions such as “Drip creation” and “Management of patient information” cannot be taken. In addition, when the work place is the “otolaryngology examination room”, “walking / moving”, “document work”, and “ME device operation” are action groups that the subject can take. It can be seen that other actions such as “sanitary treatment” cannot be taken. In this way, it can be confirmed that the work actions that the subject can take are generally limited at the work place.

  Here, in behavior identification using an acceleration sensor, it is also important to evaluate the mounting position of the acceleration sensor and the data sampling frequency at the time of data collection. For example, as a study of attaching acceleration sensors to multiple positions (parts) of the subject's body and evaluating the relationship with the recognition rate for the positions, “L.Bao and SS Insille ,: Activity recognition from user-annotated acceleration data, in Proc. of Pervasive 2004, vol. LNCS 3001, A. Ferscha and F. Mattern (Eds.), Springer-Verlag, Berlin Heidelberg, pp. 1-17, 2004. (Reference 1) “N. Kern, B. Schiele, and A. Schmidt,“ Multi-sensor activity context detection for wearable computing, ”in European Symposium on Ambient Intelligence (EUSAI), Eindhoven, The Netherlands, Nov. 2003.” (reference 2 ). In the former, the mounting position is evaluated using acceleration sensors that are mounted asymmetrically on five positions of the body, and about 80% of identification results are obtained with respect to identification of 20 kinds of daily activities in only two positions of the right wrist and left thigh. Have gained. In the latter, the optimum mounting position of the acceleration sensor is evaluated by changing the mounting position of the acceleration sensor used for identification for each action to be identified and dividing the upper body and the lower body.

  There are also various studies that evaluate the relationship between the sampling frequency and the identification rate when acquiring acceleration data. In many studies, data is collected at a sampling frequency of 50 Hz or higher (references 1 and 3). However, Reference 3 is “N. Ravi, N. Dandekar, P. Mysore, and ML Littman,: Activity recognition from accelerometer data, American association for artificial intelligence (www.aaai.org), July 2005.” . Also, “CV Bouten, KT Koekkoek, M. Verduin, R. Kodde, and JD Janssen,“ A triaxial Accelerometer and portable data processing unit for the assessment of daily physical activity, ”IEEE Trans. On Bio-Medical Eng., Vol. 44, no. 3, pp 136-147, 1997. (Ref. 4), it is necessary to have a sampling frequency of 20Hz or more to measure human behavior such as “walking” and “running”. There are evaluation results. On the other hand, “Akihiro Kawahara, Hiroyuki Morikawa, Yuki Aoyama,“ Context Estimation Using Small Wireless Sensors and its Applications, ”NPO Wearable Computer Research and Development Organization, Wearable Computing Research Group, Vol. 1, "No. 3, pp. 2-6, Dec. 2005." (Reference 5), a single biaxial acceleration sensor that samples at 10Hz or less is used, and a discrimination rate of nearly 100% is obtained. .

  Thus, it can be confirmed from past studies that the effective mounting position of the acceleration sensor 64 and the sampling frequency of effective acceleration data are different depending on the action to be identified.

  For this reason, in this embodiment, a plurality of motion measurement devices 60 (hereinafter sometimes referred to as “optimum sensor set”) used to identify one behavior from a group of behaviors that can be performed by the subject for each place. The sampling frequency (hereinafter sometimes referred to as “optimum parameter”) of each of the acceleration sensors 64 included in the plurality of motion measuring devices 60 to be used is obtained in advance, and the (current position) location of the subject is specified. The action of the subject is identified based on the acceleration data obtained according to the optimum sensor set and the optimum parameter corresponding to the specified place.

  Hereinafter, a method for determining an optimum sensor set and an optimum parameter (constructing an optimum sensor parameter DB 18) will be described, and a method for identifying a subject's action using the optimum sensor parameter DB 18, that is, the decided optimum sensor set and optimum parameters will be described. I will do it. As described above, it is assumed that the behavior of the subject is monitored in advance, and the acceleration data with the behavior label attached is stored in the sensor DB 14 in time series for each location.

  Briefly, a feature amount (feature vector) is extracted from acceleration data for an action group corresponding to a certain place, and the extracted feature vector is subjected to principal component analysis. Then, as a result of the principal component analysis, a feature amount (element) having a large contribution rate is selected, and the motion measurement device 60 that does not have the selected feature amount uses a set of motion measurement devices 60 (sensor set). Excluded from. However, when there is no acceleration sensor 64 (motion measurement device 60) that does not have the selected feature amount, the identification information of all the motion measurement devices 60 (60a-60f) is obtained as the optimum sensor set at the location. Is acquired.

  When the motion measuring device 60 that does not have the selected feature quantity is excluded, a feature vector is extracted only from acceleration data from the motion measuring device 60 (acceleration sensor 64) excluding the excluded motion measuring device 60. And principal component analysis. Then, a feature amount having a large contribution rate is selected. Such a process is repeated until there is no motion measurement device 60 that does not have the selected feature amount, that is, the motion measurement device 60 to be excluded. Then, the combination of the motion measurement devices 60 that has not been excluded is stored in the optimum sensor parameter DB 18 as the optimum sensor set at the location.

  When the optimum sensor set for each location is determined, the sampling frequency is changed for each of the acceleration sensors 64 of the plurality of motion measuring devices 60 included in each optimum sensor set, and the acceleration data is resampled. Next, feature amounts are extracted from each of the resampled acceleration data by changing the sampling frequency. Then, an identification error for each extracted feature quantity is calculated, and the sampling frequency set for each of the acceleration sensors 64 when the identification error is minimized is stored in the optimum sensor parameter DB 18 as the optimum parameter at that location. .

  Then, the server 12 detects the subject and its current position based on the detection data from the passage sensor 80, reads out the optimum sensor set and the optimum parameter corresponding to the place including the current position from the optimum sensor parameter DB 18, and the subject Is transmitted to the repeater 40 attached. Therefore, in the repeater 40, the motion measurement device 60 to be used is determined, and the sampling frequency of acceleration data in the motion measurement device 60 to be used is set. Thereafter, the server 12 detects the acceleration data transmitted from the repeater 40, extracts the feature amount from the detected acceleration data, and selects one of the action groups corresponding to the location based on the extracted feature amount. Identify actions.

  Specifically, the server 12 executes the optimum sensor parameter learning process according to the flowcharts shown in FIGS. 9 to 11, determines the optimum sensor set and the optimum parameter for each location, and constructs the optimum sensor parameter DB 18. . Moreover, the server 12 performs an action identification process according to the flowchart shown in FIG. On the other hand, the repeater 40 executes data transmission processing according to the flowchart shown in FIG.

  As shown in FIG. 9, when the optimal sensor parameter learning process is started, the server 12 reads a learning data set from the knowledge DB 16 in step S1. Specifically, the server 12 reads from the sensor DB 14 acceleration data detected by the motion measurement device 60 attached to the subject and the location of the subject as behavior records of one or more subjects. In this learning data set, an action (behavior label) corresponding to the acceleration data is given.

  Next, in step S3, the position Liε {L1, L2,..., LNL} is selected. In this position Li, labels (identification information) of all positions (places) are described. Although detailed description is omitted, the position Li is determined in advance by an administrator or user of the behavior identification system 10 and stored in the knowledge DB 16.

In step S5, it is determined whether the variable i exceeds the maximum value NL (i> NL). That is, it is determined whether or not an action identification sensor set S ^* Li and an optimum parameter Θ ^* Li described later have been obtained for all positions included in the position Li.

  If “YES” in the step S5, that is, if the variable i exceeds the maximum value NL, the optimum sensor parameter learning process is ended as it is. On the other hand, if “NO” in the step S5, that is, if the variable i is equal to or less than the maximum value NL, the action A (Li) as shown in the equation 1 is read in a step S7. Here, the action A (Li) is an action label included in an action group (behavior class) that can be taken by the subject at the position Li, and is stored in the knowledge DB 16 in advance. An example will be described with reference to FIG. 7 and FIG. 8. When the position Li is “ophthalmic examination room”, the behavior groups that the subject can take are “walking / moving”, “document work”, “ophthalmic examination assistance”. The action A (Li) includes six types of action labels {a1, a2, a3, a4, a5, a6}, “preparation for ophthalmic examination”, “hygiene processing”, and “patient transfer”.

[Equation 1]
A (Li) = {a1, a2,..., ANLi}
In subsequent step S9, all sensor sets are read out as set S. In this embodiment, the total sensor set means the motion measuring devices 60a-60d, and the set (sensor set) S means a set of identification information about the motion measuring devices 60a-60d. As shown in FIG. 11, in the next step S11, acceleration data Z (S, A (Li)) corresponding to the read action label is read. Further, in step S13, the server 12 calculates a feature amount for determining the optimum sensor set.

  When extracting feature values from acceleration data, for example, time-series data (acceleration data) is divided into sliding windows of a fixed time, and the feature values are obtained for each window. In addition, since the method of calculating | requiring this feature-value is disclosed by said reference document 1, please refer this. The parameters of the sampling frequency of acceleration data and the number of samples in the window are set in the same manner as in Reference 2, and 50 Hz data is divided into 256 sample windows that overlap by 128 samples for each action. Therefore, one window length is 5.120 seconds. FIG. 12 shows an example of a method of dividing into sliding windows. Each window is indicated by a dotted frame. However, in FIG. 12, for the sake of clarity, only the acceleration sensor 64 in the motion measuring device 60 attached to the right arm is shown, and the adjacent windows are displayed with a slight shift in the vertical and horizontal directions. is there.

In this embodiment, five types of feature quantities of average, standard deviation, energy, frequency domain entropy, and correlation coefficient are calculated for each window based on the acceleration. These are the feature quantities used in the studies of the above Reference 1 and Reference 3. The average and standard deviation are the average and standard deviation of the acceleration of each axis, respectively. The energy was obtained as the sum of the square of the amplitude by performing FFT (Fast Fourier Transform) on the acceleration of each axis. However, the DC component and the folded part are excluded. When the FFT amplitude components of the acceleration of one axis are F ₁ , F ₂ , F ₃ ,..., The energy is obtained by Equation 2. Here, w is the number of samples in the window (256 in this embodiment).

[Equation 2]

The frequency domain entropy (hereinafter sometimes simply referred to as “entropy”) is obtained based on the probability distribution p. As shown in Equation 3, the probability distribution p is obtained by normalizing the FFT component F _i excluding the DC component and the folded portion with the sum of all components.

[Equation 3]

  Thus, entropy is obtained according to Equation 4.

[Equation 4]

  The correlation coefficient is a correlation coefficient related to acceleration between two axes. At this time, the target two axes include not only combinations within one acceleration sensor 32 but also combinations straddling arbitrary two acceleration sensors 32. When (x, y) is a vector having an acceleration of a certain axis, cov (x, y) is a covariance of x and y, and σx and σy are standard deviations of x and y, respectively, between xy The correlation coefficient corr (x, y) is defined by Equation 5.

[Equation 5]

  The feature quantity calculated in this way becomes each element of the feature vector X, and a k-dimensional feature vector X as shown in Equation 6 is obtained.

[Equation 6]
X = {X1, X2,..., Xk}
Returning to FIG. 10, in the subsequent step S <b> 15, the feature vector X is subjected to principal component analysis, and a feature having a contribution ratio larger than the threshold τ is selected. This selected feature vector Xr is expressed by Equation 7.

[Equation 7]
Xr = {Xr1, Xr2,... Xrk} (rk <k)
As can be seen from Equation 7, the feature vector Xr is represented by the selected rk feature quantities. Here, as described above, there are five types of feature amounts, but the feature amount is obtained for each axis of acceleration, and the correlation coefficient is obtained for all combinations between two different axes. Thus, the feature vector X is obtained. Therefore, for example, when using the acceleration detected by the four acceleration sensors 64, 12 (3 axes × 4) pieces of “average”, “standard deviation”, “energy”, and “entropy” are used. A feature amount is obtained, and 66 ( ₁₂ C ₂ = 12 × 11/2) feature amounts are obtained as correlation coefficients. That is, the feature vector X in this case has 114 dimensions. At this time, the variable k = 114. However, the correlation coefficient obtains the correlation between the acceleration of each axis (the acceleration of 12 axes) of the four acceleration sensors 64 and the acceleration of the other axes.

  When the principal component analysis is performed on the feature vector X, eigenvalues λi (i = 1, 2,..., K) from the first principal component to the kth principal component are calculated in descending order of the eigenvalues. At this time, the contribution rate is defined as a value (percentage) obtained by dividing each eigenvalue by the total value of the eigenvalues. That is, the contribution ratio of each eigenvalue λi is expressed by Equation 8.

[Equation 8]

  The feature vector Xr is obtained by determining whether or not the value of the contribution rate exceeds the threshold value η. For example, the threshold η is determined to be 1 (%). The greater the eigenvalue λi, the greater the variance of the principal component scores, and the greater the eigenvector (weighted composite feature vector of the original feature axis) corresponding to (belonging to) each eigenvalue λi is to explain the entire feature space (the amount of information is large). ). However, it can be said that the feature axis having an extremely small eigenvalue λi is not necessary because the force explaining the entire feature space is small. Therefore, by deleting the feature axis (feature amount) having a small eigenvalue λi from the feature vector X, the feature vector Xr is compressed to the feature vector Xr with less dimensions.

  In this embodiment, the feature vector Xr is obtained using the contribution rate, but the cumulative contribution rate may be used. Here, the cumulative contribution rate means the contribution rate accumulated in order from the first principal component. In this case, for example, the threshold value η is determined to be 95 (%), and a feature vector Xr that employs the first principal component to the smallest mk principal component (mk ≦ k) exceeding the threshold η is generated.

  Returning to FIG. 10, in the next step S17, a sensor set Sm not used for the feature vector Xr is obtained. That is, the motion measuring device 60 (acceleration sensor 64) that has an element in the feature vector X but does not have an element in the feature vector Xr is detected. In step S19, it is determined whether the sensor set Sm is an empty set φ.

If “NO” in the step S19, that is, if the sensor set Sm is not the empty set φ, the sensor set S is updated in a step S21 (S = S−Sm). That is, the number of motion measurement devices 60 used for action identification is reduced. Then, it returns to step S11. On the other hand, if “YES” in the step S19, that is, if the sensor set Sm is an empty set φ, the sensor set S is determined as the optimum sensor set S ^* Li in a step S23 as shown in FIG.

  Subsequently, in step S25, a parameter set ΘLi is selected from the sensor parameter space. Here, as described above, the parameter set ΘLi includes all combinations of sampling frequencies set in the acceleration sensor 64 of the motion measuring apparatus 60 to be used. In this embodiment, the sampling frequency is changed in five stages of 50 Hz, 25 Hz, 12.5 Hz, 6.25 Hz, and 3.125 Hz for each of the acceleration sensors 64 of the plurality of motion measuring devices 60 indicated by the optimum sensor set. All combinations are included.

In the subsequent step S27, the acceleration data of the optimal sensor set S ^* Li is resampled with the selected parameter set ΘLi. In step S29, a feature is extracted from each of the acceleration data of the resampled optimum sensor set S ^* Li. Each element (feature amount) of the feature vector is the same as the feature vector Xr described above. When performing feature calculation, the number of samples included in the window differs depending on the sampling frequency, but the number of samples is variable by keeping the window size length constant.

  Next, in step S31, an identification error is calculated using the classifier C. As for the action label identification method, supervised learning is applied to each action label using the sliding window feature amount extracted in step S29. For example, although illustration is omitted, the feature quantity at each sampling frequency is mapped to the feature space, and a distribution (cluster) is formed on the feature space. The cluster is classified into each action label by a discriminator such as a well-known SVM (Support Vector Machine).

  When constructing a multi-class SVM, training is performed by training a pair-wise discriminator for discriminating each class from other individual classes for the number of classes. In this embodiment, an SMO (Sequential Minimal Optimization) method is used, and a one-dimensional polynomial kernel is used as a kernel function. However, regarding the SMO method, “SS Keerthi, S. K, Shevade, C. Bhattacharyya, KR K, Murthy: Improvements to Platt's SMO Algorithm for SVM Classifier Design, Neural Computation, 13 (3), pp. 637-649, 2001 "(reference 7).

In a succeeding step S33, it is determined whether or not the entire parameter space has been searched. If “NO” in the step S33, that is, if the entire parameter space is not searched, the process returns to the step S25 to select another parameter set ΘLi from the sensor parameter space. On the other hand, if “YES” in the step S33, that is, if the entire parameter space is searched, the parameter set ΘLi that minimizes the identification error is determined as the optimum parameter Θ ^* Li as it is in a step S35.

In step S37, the optimum sensor set S ^* Li and the optimum parameter Θ ^* Li at the position Li are stored in the optimum sensor parameter DB 18, and the process returns to step 3 shown in FIG. That is, in step S37, the server 12 stores the optimum sensor set S ^* Li and the optimum parameter Θ ^* Li in the optimum sensor parameter DB 18 in association with each position Li.

  FIG. 13 is a flowchart showing the action identification process of the server 12 as described above. As shown in FIG. 13, when starting the action identification process, the server 12 receives the detection data (light receiving module ID and transmitter ID) transmitted from the passage sensor 80 in step S51, and refers to the knowledge DB 16. Identify the subject and the location (current position) where the subject exists.

In step S53, the optimum sensor set S ^* Li and the optimum parameter Θ ^* Li corresponding to the detected position Li of the subject are read from the optimum sensor parameter DB 18. In the following step S55, the optimum sensor set S ^* Li and the optimum parameter Θ ^* Li are transmitted to the repeater 40.

Next, in step S57, it is determined whether or not the acceleration data from the repeater 40 that has transmitted the optimum sensor set S ^* Li and the optimum parameter Θ ^* Li in step S55 has been received. As described above, the acceleration data transmitted from the repeater 40 is detected by each acceleration sensor 64 of the motion measuring device 60 included in the optimum sensor set S ^* Li, for example, at a sampling frequency according to the optimum parameter Θ ^* Li. Acceleration data in the case. If “NO” in the step S57, that is, if the acceleration data is not received, the process waits until the acceleration data is transmitted from the repeater 40. On the other hand, if “YES” in the step S57, that is, if acceleration data is received, a feature is extracted from the acceleration data as it is in a step S59.

  In step S59, features are extracted using the acceleration data received in step S57, and a feature vector Xa as shown in Equation 9 is obtained.

[Equation 9]
Xa = {Xa1, Xa2,..., XaM}
Next, in step S61, action identification processing is executed (ωa = classify_action (Xa)). For example, for each behavior class, data (mapping data) obtained by mapping a feature amount (here, “reference behavior feature amount” for convenience of explanation) for each behavior label included in the behavior class into a feature space is used. Store in advance in the knowledge DB 16. Further, a class classifier for identifying an action class corresponding to the position Li is learned, and using these, the feature amount calculated this time (here, for the convenience of explanation, “the current feature feature amount” ") Is identified (specified) as the subject's behavior.

  In subsequent step S63, the action identification result ωa is output. For example, the action identification result ωa is displayed as text on a monitor (not shown) connected to the server 12, or is output as a voice to a speaker (not shown) also connected to the server 12. Or output by means. However, the action identification result ωa may be transmitted (output) by e-mail to another terminal (computer or telephone). In addition to simply outputting the action identification result ωa, the acceleration data detected this time may be labeled.

  In the next step S65, the identification result (position and action) is stored in the action record DB 20. In step S67, it is determined whether or not to end the action identification process. Here, it is determined whether or not an end instruction from the user has been input, or processing for all desired acceleration data has been completed. If “NO” in the step S67, that is, if the action identifying process is not ended, the process returns to the step S51 as it is, the position Li of the subject is detected with reference to the ID newly received from the passage sensor 80, and the action identifying process is performed. Execute. On the other hand, if “YES” in the step S67, that is, if the action identification process is ended, the action identification process is ended as it is.

FIG. 14 is a flowchart showing the data transmission process of the repeater 40 as described above. As shown in FIG. 14, when the data transmission process is started, the repeater 40 determines whether or not the optimal sensor set and the optimal parameter are received from the server 12 in step S81. The optimum sensor set and optimum parameters received from the server 12 are the optimum sensor set S ^* Li and the optimum parameter Θ corresponding to the subject position Li transmitted from the server 12 in S55 shown in FIG. ^* Li.

If “NO” in the step S81, that is, if the optimum sensor set S ^* Li and the optimum parameter Θ ^* Li are not received, the process proceeds to a step S87 as it is. That is, in such a case, the motion measuring device 60 (acceleration sensor 64) to be used and its sampling frequency are not set (changed).

On the other hand, if “YES” in the step S81, that is, if the optimum sensor set S ^* Li and the optimum parameter Θ ^* Li are received, the acceleration data of the motion measuring device 60 not included in the optimum sensor set is transmitted in a step S83. To pause. For example, the repeater 40 transmits a command (pause command) for pausing the transmission of acceleration data to the motion measurement device 60 not included in the optimal sensor set, and the acceleration data to the motion assumption device 60 included in the optimal sensor set. A command (start command) for starting (or resuming) transmission of is transmitted. Therefore, when receiving the pause command, the CPU 62 of the motion measuring device 60 pauses the transmission of acceleration data until the next start command is received. However, when the pause command is received, the motion measuring device 60 may be operated in the sniff mode. Here, the sniff mode is a low power consumption mode peculiar to Bluetooth that performs transmission / reception only at a constant time interval (sniff interval) and suppresses power consumption during other times.

In step S83, the repeater 40 causes the motion measurement device 60 not included in the optimum sensor set S ^* Li to stop transmitting acceleration data. Specifically, the motion measuring device 60 is switched to the sniff mode. The sniff mode is a low power consumption mode peculiar to Bluetooth that performs transmission / reception only at a constant time interval (sniff interval) and suppresses power consumption during other times.

In the subsequent step S85, the repeater 40 sets the parameters of the motion measuring device 60 included in the optimal sensor set S ^* Li as the optimal parameters. Specifically, the repeater 40 sets the sampling frequency of the motion measurement device 60 (acceleration sensor 64) included in the optimum sensor set S ^* Li to the sampling frequency indicated by the optimum parameter Θ ^* Li.

  In step S87, acceleration data is acquired. Here, acceleration data is detected from the acceleration sensor 64 included in the motion measuring device 60 to be used, and the sampling frequency is set. When the optimum sensor and the optimum parameter are not set, acceleration data detected at a predetermined sampling frequency (for example, 50 Hz) is acquired from the acceleration sensors 64 of all the motion measurement devices 60.

  In a succeeding step S89, it is determined whether or not a certain time (for example, the length of time corresponding to the sliding window described above) has elapsed, that is, whether or not a sufficient amount of acceleration data has been acquired to identify the action. In addition, although illustration is abbreviate | omitted, the elapsed time after starting acquisition of acceleration data is counted by the internal timer of the repeater 40. If “NO” in the step S89, that is, if a predetermined time has not elapsed, the process returns to the same step S89. On the other hand, if “YES” in the step S89, that is, if a certain time elapses, the acceleration data for the certain time is transmitted to the server 12 in a step S91, and the process returns to the step S81.

  According to this embodiment, the action is identified based on the acceleration data obtained from the plurality of motion measurement devices attached to the subject according to the optimum sensor set and the optimum parameter determined according to the place where the subject exists. The complex behavior of the subject can be identified more accurately.

  Further, according to this embodiment, the repeater 40 causes the motion measurement device 60 that is not included in the optimum sensor set to stop transmitting acceleration data according to the current position of the subject. Thereby, useless power consumption of the motion measuring device 60 can be eliminated. Therefore, the life of the battery to be used or the driving time by one charge can be made as long as possible.

  Similarly, since the sampling frequency of the acceleration sensor is set to an optimal value depending on the location, the power consumption of the acceleration sensor (motion measurement device) is higher than when all acceleration sensors are always driven at a high frequency. Can be suppressed.

  In this embodiment, the optimum sensor set and the optimum parameter associated with each place are stored in the optimum sensor parameter DB 18, and according to both the optimum sensor set and the optimum parameter determined according to the place where the subject exists. Although the behavior is identified based on the acceleration data obtained from the plurality of motion measurement devices 60 attached to the subject, it is not necessary to be limited to this. For example, only the optimum sensor set associated with each place is stored in the optimum sensor parameter DB 18 and only the optimum sensor set determined according to the place where the subject exists is determined. Therefore, the motion measuring device 60 included in the optimum sensor set is used. The behavior can also be identified based on the acceleration data obtained from. Similarly, only the optimum parameters associated with each location are stored in the optimum sensor parameter DB 18, and all the exercises attached to the subject according to only the optimum parameters determined according to the location where the subject exists. It is also possible to set a parameter of the measuring device 60 as an optimum parameter and identify an action based on acceleration data obtained from the motion measuring device 60.

  In this embodiment, the server 12 detects the location of the subject and identifies the behavior of the subject. However, the present invention is not limited to this. If the knowledge DB 16 and the optimum sensor parameter DB 18 are provided in the repeater 40, the repeater 40 can detect the subject's location and identify the subject's behavior. In such a case, the action can be identified regardless of whether the server 12 and the repeater 40 are online or offline.

  Furthermore, in this embodiment, the acceleration data is transmitted from the repeater 40 to the server 12 at regular intervals, but may be transmitted to the server 12 after all the acceleration data is acquired. Here, all the acceleration data includes acceleration data for the behavior of the subject in one day, acceleration data that can be stored in the memory of the repeater 40, or a battery (primary battery, secondary battery) that drives the repeater 40. ) Means the amount of acceleration data that can be recorded.

  In this embodiment, the repeater 40 pauses the transmission of the acceleration data from the motion measurement device 60 that is not included in the optimum sensor set. However, the present invention is not limited to this. For example, the repeater 40 may control the on / off of the motion measuring device 60 itself that is not included in the sensor set. Alternatively, the repeater 40 is configured not to receive acceleration data from the motion measurement device 60 that is not included in the sensor set, or to receive acceleration data from all the motion measurement devices 60, and to be included in the sensor set. Only the acceleration data from 60 may be transmitted to the server.

  Furthermore, in this embodiment, the acceleration data is transmitted from the motion measurement device 60 to the server 12 via the repeater 40. However, the acceleration data is transmitted directly from the motion measurement sensor 60 to the server 12 without using the repeater 40. Data may be transmitted.

  In this embodiment, the subject is provided with one transmitter 100 and the passage sensor 80 having the light receiving module is provided in various places. However, the present invention is not limited to this. For example, the subject may be provided with one light receiving module, and the passage sensor 80 including the transmitter 100 may be arranged in various places. In such a case, since the light receiving module of the subject receives the transmitter ID from the transmitter 100, if the database storing the location corresponding to the transmitter ID is provided in the repeater 40, the repeater 40 uses the subject. Can be identified. As a matter of course, the detected transmitter ID can be transmitted to the server 12 so that the server 12 can specify the location of the subject.

  Furthermore, in this embodiment, the acceleration sensor 64 is used to detect the behavior (motion) of the subject, but it is not necessary to be limited to this. For example, other sensors such as a gyro sensor, an illuminance sensor, or a temperature sensor can be used instead of the acceleration sensor. For example, when a gyro sensor (three axes) is used, angular velocity data from the gyro sensor is acquired as operation data.

  Furthermore, in this embodiment, SVM (Support Vector Machine) is adopted as the classifier, but it is not limited to this. For example, a well-known NN (Nearest Neighbor), C4.5, k-nearest neighbor method, etc. can be adopted.

  Further, in this embodiment, a case has been described in which an action for one subject is identified. However, there may be a plurality of subjects. In such a case, since the behavior (behavior) is different for each subject and the optimum sensor set and the optimum parameter are considered to be different due to differences in height and body shape, the optimum sensor parameter DB 18 is created for each subject. What is necessary is just to make it identify action using optimal sensor parameter DB18 corresponding to a test subject. Moreover, if the average of acceleration data of each subject is taken, an optimum sensor set and optimum parameters that can be used for all subjects can be created.

  Furthermore, in this embodiment, the motion measuring device 60 is attached to both upper arms, chest, and lower back to identify the behavior, but the number of acceleration sensors to be used and the mounting positions are applied to the behavior identification system 10 ( Use) Varies depending on the environment. Also, the behavior to be identified varies depending on the usage environment of the behavior identification system 10 and the like.

  In this embodiment, the action identification system 10 is provided in a hospital and a subject is a nurse. However, the present invention is not limited to this. For example, the present invention can be applied to any environment as long as the environment tends to relate work behavior and work place.

FIG. 1 is an illustrative view showing one example of an action identification system of the present invention. FIG. 2 is an illustrative view showing an electrical configuration of the repeater of FIG. FIG. 3 is an illustrative view showing a state in which a subject wears the repeater and the motion measurement device of FIG. 4 is an illustrative view showing an electrical configuration of the motion measuring apparatus of FIG. FIG. 5 is an illustrative view showing an electrical configuration of the passage sensor of FIG. FIG. 6 is an illustrative view for explaining the appearance of the passage sensor of FIG. FIG. 7 is an illustrative view for explaining the arrangement state of the passage sensors. FIG. 8 is a graph showing the probability of work activity occurring at a work place. FIG. 9 is a flowchart showing a part of the optimum sensor parameter learning process of the server shown in FIG. 10 is another part of the optimum sensor parameter learning process of the server shown in FIG. 1, and is a flowchart subsequent to FIG. FIG. 11 is another part of the optimum sensor parameter learning process of the server shown in FIG. 1, and is a flowchart subsequent to FIG. FIG. 12 is an illustrative view showing an example in which acceleration data is divided by a sliding window of a fixed time. FIG. 13 is a flowchart showing the action identification processing of the server shown in FIG. FIG. 14 is a flowchart showing data transmission processing of the repeater shown in FIG.

Explanation of symbols

10 ... Action identification system 12 ... Server 14 ... Sensor DB
16… Knowledge DB
18 ... Optimal sensor parameter DB
20 ... Action record DB
DESCRIPTION OF SYMBOLS 30 ... Network 40 ... Repeater 60 ... Motion measuring apparatus 80 ... Passing sensor 100 ... Transmitter

Claims (10)

An action identification system for identifying one action that a subject can perform from a plurality of action groups that differ depending on a place,
Among the plurality of motion measurement devices mounted on different parts of the subject, a storage unit that stores an optimal combination of the motion measurement devices in association with each location,
Position detecting means for detecting the current position of the subject;
Reading means for reading out the combination of the motion measurement devices corresponding to the location including the current position of the subject detected by the position detection means from the storage means as an optimal sensor set,
Motion detection means for detecting outputs from a plurality of motion measurement devices included in the optimum sensor set read by the reading means;
A behavior identification system comprising behavior identification means for identifying the behavior of the subject from the detection result of the motion detection means. An action identification system for identifying one action that a subject can perform from a plurality of action groups that differ depending on a place,
Storage means for storing optimum parameters set in each of a plurality of motion measurement devices mounted on different parts of the subject in association with each location;
Position detecting means for detecting the current position of the subject;
Reading means for reading out the optimum parameter corresponding to the location including the current position of the subject detected by the position detecting means from the storage means,
Parameter setting means for setting the optimum parameters read by the reading means in each of the plurality of motion measuring devices;
Motion detection means for detecting outputs from the plurality of motion measurement devices whose parameters are set by the parameter setting means;
A behavior identification system comprising behavior identification means for identifying the behavior of the subject from the detection result of the motion detection means. An action identification system for identifying one action that a subject can perform from a plurality of action groups that differ depending on a place,
A plurality of motion measurement devices mounted on different parts of the subject, a first storage unit that stores an optimal combination of motion measurement devices in association with each location;
Second storage means for storing optimum parameters set in each of a plurality of motion measurement devices included in the combination of the optimal motion measurement devices in association with each location;
Position detecting means for detecting the current position of the subject;
First reading means for reading out a combination of motion measuring devices corresponding to a place including the current position of the subject detected by the position detecting means from the first storage means as an optimum sensor set;
A second reading means for reading the optimum parameter set in each of the plurality of motion measuring devices included in the optimum sensor set read by the first reading means from the second storage means;
Parameter setting means for setting the optimum parameter read by the second reading means to each of the plurality of motion measurement devices included in the optimum sensor set;
Motion detection means for detecting outputs from a plurality of motion measurement devices included in the optimum sensor set;
A behavior identification system comprising behavior identification means for identifying the behavior of the subject from the detection result of the motion detection means.   The action identification system according to claim 1, further comprising a pause unit that pauses a motion measurement device not included in the optimum sensor set. A learning means for learning a class classifier by associating each behavior with a reference behavior feature amount acquired when the subject performs a plurality of behaviors in advance,
The action identifying means includes
Extraction means for extracting the behavior feature of the subject from the detection result of the motion detection means;
The action identification according to claim 1, further comprising action specifying means for specifying the action of the subject from the action feature amount extracted by the extracting means using the class classifier learned by the learning means. system. Among the plurality of motion measurement devices attached to different parts of the subject, the computer comprises a storage means for storing an optimum combination of motion measurement devices in association with each location, and is a computer behavior identification method for identifying the subject's behavior. There,
(a) detecting the current position of the subject;
(b) Read the optimal combination of motion measurement devices corresponding to the location including the current position of the subject detected in step (a) from the storage means as an optimal sensor set;
(c) detecting outputs from a plurality of motion measuring devices included in the optimum sensor set read in step (b); and
(d) A behavior identification method for identifying the behavior of the subject from the detection result of the step (c). A computer behavior identification method comprising storage means for storing optimal parameters set in each of a plurality of motion measurement devices mounted on different parts of a subject in association with each location, and identifying the subject's behavior ,
(a) detecting the current position of the subject;
(b) Read the optimum parameter corresponding to the location including the current position of the subject detected in step (a) from the storage means,
(c) setting the optimum parameters read in step (b) to each of the plurality of motion measuring devices;
(d) detecting outputs from the plurality of motion measuring devices; and
(e) A behavior identification method for identifying the behavior of the subject from the detection result of the step (d). Among a plurality of motion measurement devices mounted on different parts of the subject, a plurality of first measurement means for storing a combination of optimal motion measurement devices in association with each location and a plurality included in the combination of the optimal motion measurement devices A second storage means for storing the optimum parameters set in each of the exercise measuring devices in association with each location, and a computer behavior identification method for identifying the subject's behavior,
(a) detecting the current position of the subject;
(b) Read an optimal combination of motion measurement devices corresponding to the location including the current position of the subject detected in step (a) from the first storage means as an optimal sensor set;
(c) Read the optimum parameters in the plurality of motion measurement devices included in the optimum sensor set read in step (b) from the second storage means as the optimum parameters;
(d) setting the optimum parameter read in step (c) to each of the plurality of motion measurement devices included in the optimum sensor set;
(e) detecting outputs from a plurality of motion measurement devices included in the optimum sensor set for which parameters are set in the step (d); and
(f) A behavior identification method for identifying the behavior of the subject from the detection result of the step (e). An optimal sensor set determination method for determining an optimal combination of motion measurement devices associated with each location from a plurality of motion measurement devices attached to different parts of the subject in order to identify the subject's behavior,
(a) extracting a first feature amount from motion data corresponding to a group of actions performed by the subject at a certain location;
(b) Based on the principal component analysis of the first feature amount extracted in the step (a), a second feature amount obtained by extracting an element having a contribution rate larger than a predetermined threshold from the first feature amount is selected. ,
(c) excluding the motion measurement device that does not have the second feature value selected in step (b);
(d) Steps (a) to (c) are repeated until there are no more exclusion targets in step (c), and (e) a combination of motion measurement devices that have not been excluded is determined as an optimal sensor set. An optimal sensor set determination method. Optimal parameter determination that determines the optimal parameter associated with each location set in the motion detection sensor provided in each of the plurality of motion measurement devices attached to different parts of the subject in order to identify the subject's behavior A method,
(a) With respect to motion data detected by a plurality of motion detection sensors set to a predetermined sampling frequency and corresponding to a group of behaviors performed by the subject at a certain location, the sampling frequency of each of the plurality of motion detection sensors is set to Change and resample,
(b) extracting a feature quantity from the motion data resampled in the step (a),
(c) calculating an identification error for the feature amount extracted in step (b),
(d) repeating steps (a) through (c) until the identification error is calculated for all sampling frequencies within a predetermined range; and
(e) An optimum parameter determination method for determining a combination of sampling frequencies as an optimum parameter that minimizes the identification error calculated in the step (c).

Images