JP4950865B2 - Sensor node control system and method - Google Patents

Description

  The present invention relates to a sensor node control system and method for controlling a plurality of sensor nodes having a sensor function and capable of communicating with each other.

  In recent years, it has become possible to measure and estimate in real time the results of measurements of body temperature and movement in animals (individuals) such as cows and horses by attaching a predetermined sensor to the target individual. It has become. Furthermore, not only the individual to be measured, but also the communication radio waves emitted by sensors attached to other individuals in the vicinity of the individual, and the value measured by the nearby sensor is used to Some perform state estimation more accurately (see Patent Document 1).

JP 2006-320290 A R.O.Duda, P.E.Hart, and D.G.Stork, Director: Morio Onoe, “Pattern Classification Second Edition”, John Wiley & Sons, New Technology Communications, July 3, 2001, first edition.

  By the way, what is called a wearable sensor or a user sensor node used by being attached to an individual as described above generally cannot use a commercial power supply, but is driven by a battery. By using a battery, it is possible to reduce the size and carry it, and to measure the state of a moving individual. However, when driven by a battery, an operation that requires a larger amount of power than acquiring a large amount of high-definition data and communicating immediately may accelerate battery consumption and may not be practical.

  For this reason, in the sensor node driven by the battery, the operation time of the sensor is limited, but there is a possibility that necessary information is lost in the thinned measurement data. Further, the method of accumulating data for a certain period and transmitting it in a batch cannot satisfy immediacy. As described above, the conventional sensor node using a battery has a problem that there is a limit to instant communication of more effective data.

  The present invention has been made to solve the above-described problems, and an object thereof is to enable immediate communication of more effective data in a sensor node driven using a battery. To do.

  The sensor node control system according to the present invention is provided with an object sensor unit that detects an operation received by an object and transmits an object ID for identifying itself, and measures the state of the user attached to the user. A user sensor unit that includes a state measurement unit that transmits a user state measured by the state measurement unit, an object ID transmitted from the object sensor unit, and a user behavior state corresponding to the user state transmitted from the user sensor unit The control state of the user behavior storage unit in which a plurality of user behaviors are stored and the control state of the state measurement unit corresponding to the user behavior are changed within a change range in which the power consumption of the state measurement unit is equal to or less than a predetermined value, and the control state is changed A plurality of changed user actions estimated based on a plurality of changed user states obtained by changing the user state corresponding to the A control data generation unit that generates and generates a control state in which a change user action with the highest recognition rate indicating a rate is estimated in a change range, and controls the user sensor unit with the control state generated by the control data generation unit A user sensor control unit is provided at least.

  In the sensor node control system, the control data generation unit obtains a recognition rate by pattern recognition of the changed user state for a plurality of user actions stored in the user action storage unit. Further, the control data generation unit generates a control state by giving a weight to the recognition rate for each corresponding user action.

  In addition, the sensor node control method according to the present invention includes an object ID for identifying an object sensor unit transmitted from an object sensor unit provided on an object to detect an operation received by the object, and a user wearing A first step of storing a plurality of user behaviors indicating a user behavior state corresponding to the user status measured by the status measurement unit of the user sensor unit transmitted from the user sensor unit, and a status corresponding to the user behavior Change the control state of the measurement unit within the change range where the power consumption of the state measurement unit is less than the default value, and change the user state corresponding to the changed control state based on multiple changed user states A plurality of estimated changed user behaviors are obtained, and a control state in which the changed user behavior that maximizes the recognition rate indicating the matching rate between the obtained changed user behavior and the user behavior is estimated is obtained in the change range. A second step of, in a control state generated in the second step, at least comprising the method and a third step of controlling the user sensor unit.

  In the sensor node control method, in the second step, the recognition rate is obtained by pattern recognition of the changed user state for a plurality of user actions stored in the user action storage unit. In the second step, the recognition state is weighted for each corresponding user action to generate a control state.

  As described above, in the present invention, the control state of the state measurement unit corresponding to the user behavior is changed within a change range in which the power consumption of the state measurement unit is a predetermined value or less, and the user state corresponding to the changed control state A change user action that obtains a plurality of change user actions estimated based on a plurality of change user states obtained by changing the user ID, and that shows a matching rate between the obtained change user action and the user action is maximized. The estimated control state is determined and generated within the change range. As a result, according to the present invention, more effective data can be immediately communicated in the sensor node driven using the battery.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a configuration diagram illustrating a configuration example of a sensor node control system according to an embodiment of the present invention. This sensor node control system includes a sensor node control device 100, a user sensor unit 120 serving as a sensor node to be controlled, and object sensor units 130A, 130B, and 130C serving as other sensor nodes. The user sensor unit 120 is worn by a user who is an estimation target such as an action. The object sensor unit is attached to an object arranged around the user. FIG. 1 shows a case where three object sensor units 130 are provided. However, the present invention is not limited to this, and four or more object sensor units 130 may be provided. Also good.

  The sensor node control device 100 constituting the sensor node control system includes a user behavior state estimation unit 101, a state storage unit 102, a transmission / reception unit 103, a reception data storage unit 104, a user behavior storage unit 105, and a control data generation unit 106. , A control data storage unit 107, a user sensor control unit 108, a communication control unit 109, and an input unit 110. The user sensor unit 120 includes a user state measurement unit 121, a storage unit 122, a transmission / reception unit 123, and a control unit 124. The object sensor unit 130 </ b> A includes an operation state detection unit 131, a storage unit 132, a transmission / reception unit 133, and a control unit 134. The object sensor unit 130B and the object sensor unit 130C have the same configuration as the object sensor unit 130A.

  Next, each configuration will be described in more detail.

  First, the transmission / reception unit 103 stores data received by wireless communication in the reception data storage unit 104. In addition, the transmission / reception unit 103 transmits an instruction to transmit necessary data under the control of the communication control unit 109, receives the data transmitted as a result of this instruction, and stores it in the received data storage unit 104. From the data stored in the reception data storage unit 104 in this way, the user behavior state estimation unit 101 generates data indicating the user state (user state) and data indicating the object that has been operated (object ID). Take out.

  When each data is extracted as described above, the user behavior state estimation unit 101 estimates the state of the environment of the user including the user wearing the user sensor unit 120 from the extracted user state and object ID. To do. The user behavior state estimation unit 101 estimates the user behavior state by comparing the user state and the object ID with the user behavior result (reference data) stored in the state storage unit 102. The user action result is input from the input unit 110 by the operation of the operator and stored in the state storage unit 102 by the user action state estimation unit 101.

  For example, the reference data in which the object ID of the object in which the user state and the operation state measured by the user sensor unit 120 are detected by the object sensor unit 130A and the corresponding user action is associated with the state storage unit 102 Is remembered. With respect to the reference data stored in this way, the user behavior state estimation unit 101 estimates the user's behavior state by pattern recognition using the obtained object ID and the measured value as a feature vector. The user behavior (estimated user behavior) estimated in this way is stored in the user behavior storage unit 105.

  Next, the control data generation unit 106 changes the control state of the user state measurement unit 121 corresponding to the user behavior within a change range in which the power consumption of the user state measurement unit 121 is equal to or less than a predetermined value, and corresponds to the changed control state. Recognizing the matching rate between the obtained changed user behavior and the user behavior by obtaining a plurality of changed user behaviors estimated based on the plurality of changed user states obtained by changing the already obtained user status The control state in which the changed user action with the maximum rate is estimated is obtained and generated in the change range.

  For example, the control data generation unit 106 uses a series of user behaviors (user behavior records) stored in the user behavior storage unit 105 as training data, and determines the control state of the power source of the user state measurement unit 121 of the user sensor unit 120. Estimate the user's behavioral state when changed within the specified range, and compare these estimated behavioral state (changed user behavior) with the behavioral state (user behavior) in the training data to show these match rates Control data for controlling the power supply of the user state measurement unit 121 is generated so as to maximize the recognition rate, and the generated control data is stored in the control data storage unit 107.

  The user sensor control unit 108 uses the control data stored in the control data storage unit 107 and controls the power supply of the user state measurement unit 121 in the user sensor unit 120 by transmitting the control data, for example.

  The sensor node control device 100 is a server (computer device) including, for example, a CPU, a main storage device, an external storage device, a network connection device, and the like, and the CPU operates according to a program developed in the main storage device. Thus, each function described above is realized. Each function may be distributed among a plurality of computer devices. For example, the transmission / reception unit 103 may be connected via an interface such as a serial port, a parallel port, a LAN (Local Area Network), a USB (Universal Serial Bus), or a CF (Compact Flash) interface.

  Next, in the user sensor unit 120, the user state measurement unit 121 measures the state of the user (human body) such as the potential (muscle potential) generated by the muscles of the user's arm. The user state (user state) measured by the user state measurement unit 121 is transmitted to the sensor node control device 100 by the transmission / reception unit 123 controlled by the control unit 124. Further, the storage unit 122 stores the user sensor unit 120 and identification information for identifying the user to which the user sensor unit 120 is attached.

  For example, when the transmission / reception unit 123 receives a data transmission instruction from the sensor node control device 100, the user sensor unit 120 controls the user state measurement unit 121 together with the identification information stored in the storage unit 122 under the control of the control unit 124. Transmits the user state being measured to the sensor node control device 100.

  In the user sensor unit 120, the control unit 124 controls the power supply of the user state measurement unit 121 based on the control data received from the sensor node control device 100 (user sensor control unit 108).

  Note that the user state measurement unit 121 is not limited to the measurement of myoelectric potential, but the heart rate, pulse rate, blood flow, blood saturated oxygen concentration, electrocardiogram, skin temperature, skin of the target (wearing) user. Any sensor that measures biological information such as conductivity and brain waves may be used. The user state measurement unit 121 may be, for example, an acceleration sensor or a foot pressure sensor that measures a distribution of pressure applied to the sole of the foot.

  Further, it is desirable that the user sensor unit 120 has a form (shape) that can be worn in a state in which a sense of incongruity is reduced as much as possible, such as a state attached to clothes, a wristband type, a wristwatch type, or the like. Since the user sensor unit 120 is carried by a user and carried, the user sensor unit 120 includes a battery and is driven by the battery, although not shown. The transmission / reception unit 123 communicates with the sensor node control device 100 (transmission / reception unit 103) by wireless communication. Examples of the wireless communication include wireless LAN, specific low power wireless, weak wireless, and “ZigBee” ( Registered Trademark), “Bluetooth” (Registered Trademark), or the like may be used.

  Next, in the object sensor unit 130A, the operation state detection unit 131 detects a state (operation state) in which the object is operated by the user, for example, the user has an object provided with the object sensor unit 130A. To do. When the operation state is detected by the operation state detection unit 131, the control unit 134 sends the object ID stored in the storage unit 132 to the sensor node control device 100 by the transmission / reception unit 133. The storage unit 132 stores identification information (object ID) for identifying the object sensor unit 130A and the object on which the object sensor unit 130A is installed.

  The operation state detection unit 131 may be a contact sensor that detects a user's contact with a target object or a vibration sensor that detects vibration of the object. Further, when the target object is, for example, an electric device or an electronic device used at home, a sensor that detects the state of these operations may be used. The transmission / reception unit 133 is the same as described above, and communicates with the sensor node control device 100 (transmission / reception unit 103) by wireless communication. Examples of the wireless communication include a wireless LAN, a specific low power wireless, and a weak wireless. , “ZigBee” (registered trademark), and “Bluetooth” (registered trademark) may be used. Note that the object sensor unit 130A may be driven by a battery when an object to be installed moves. Further, the object sensor unit 130A may be driven by a commercial power source when the object to be installed is fixed and arranged in the home such as the above-described electric equipment. The object sensor unit 130B and the object sensor unit 130C are similar to the object sensor unit 130A.

  Here, reference data stored in the state storage unit 102 will be described. The state storage unit 102 stores, in advance, reference data in which the object ID of the object in which the user state and the operation state measured by the user sensor unit 120 are detected by the object sensor unit 130A and the corresponding user action are associated with each other. I remember it. As illustrated in FIG. 2, the reference data includes muscle myoelectric potential 1 (first measurement value) of a part of a user's muscle measured by the user state measurement unit 121, muscle myoelectric potential 2 of a muscle of another part (second). Measured value) and the object ID whose operation state is detected by the operation state detection unit 131 are associated with the user behavior observed when these are measured and detected.

  For example, for the value “a1” of the myoelectric potential 1, the value “a2” of the myoelectric potential 2, and the object ID “y”, the user is associated with the action of “picking the corresponding object”. For example, the object ID “y” is an identifier of the object sensor unit 130A, the object ID “z” is an identifier of the object sensor unit 130B, and the object ID “q” is an identifier of the object sensor unit 130C. In addition, the value of the myoelectric potential is obtained by analog-digital conversion of the value of the myoelectric potential sensor subjected to the smoothing filter processing, sampling at a certain sampling period, and integrating the obtained myoelectric potential data for a certain period, It can be considered as the strength of the muscles in a certain period.

  Next, generation of control data by the sensor node control system (user behavior storage unit 105, control data generation unit 106) of the present embodiment will be described in more detail. First, as shown in FIG. 3, an environment in which thirteen object sensor units A to M are arranged is considered. In this environment, five object sensor units A to E are arranged in the region α, five object sensor units F to J are arranged in the region β, and three object sensor units K to M are arranged in the region γ. . Each object sensor unit is mounted on, for example, furniture or daily items installed in each region.

  Among these, the user wearing the user sensor unit 120 acts. The user sensor unit 120 includes, for example, two myoelectric potential sensors as the user state measuring unit 121, and each myoelectric potential sensor measures a potential generated in a predetermined muscle in the user's arm. The arm movement by the user can be measured by the measurement value of the myoelectric potential sensor. Moreover, each object sensor part is provided with the contact sensor which detects the contact by a user, the human sensor which detects a user's approach, etc. as an operation state detection part.

  For example, each object sensor unit is attached to a place where a user lives every day or a daily life item picked up frequently. For example, in the example illustrated in FIG. 3, the object sensor unit D, the object sensor unit J, and the object sensor unit M are provided with human sensors that detect the approach of the user, and are attached to furniture or the like. And Further, the other object sensor is assumed to be provided with a contact sensor, and for example, is assumed to be attached to a daily product that can be used and moved by the user.

  The object ID and the user state sent from each object sensor unit and user sensor unit 120 provided as described above are received by the sensor node control device 100, and the user behavior state estimation unit 101 is based on the received data. Thus, the user behavior is estimated, and the estimated user behavior (user behavior record) is stored in the user behavior storage unit 105. The plurality of user action records stored in this manner are used as training data in the control data generation unit 106.

  Here, an example of the user behavior record stored in the user behavior storage unit 105 will be described with reference to FIG. In the example of the user action record shown in FIG. 4, first, the user detects the object sensor unit A, the object sensor unit B, the object sensor unit C, the object sensor unit F, the object sensor unit G, the object sensor unit H, and the object sensor unit I. , The object sensor unit K, and the object sensor unit L in this order are shown to have come into contact with living items to which they are attached. Further, it can be seen that the user is present in the region α by the presence or absence of a signal from the object sensor unit D (transmission of the object ID). In the region α, it can be seen that the user has touched the daily life item to which the object sensor unit A, the object sensor unit B, the object sensor unit C, and the object sensor unit F are attached.

  Further, based on the user state (two myoelectric potential measurement results) from the user sensor unit 120, for example, the user is inferred an action such as lifting a living item to which the object sensor unit B is attached to the hand. For example, a user's action to be noted is estimated based on a set of a user state and a signal (object ID) from the object sensor unit in each region surrounded by a dotted line. The user behavior is configured by combining the corresponding user state and the object ID from the object sensor unit with the behavior estimated in this manner, and one user behavior record is configured from the plurality of user behaviors. FIG. 4 shows that four actions indicated by “vvv”, “xxx”, “yyy”, and “zzz” are included in a series of actions by the user. In addition, there is a user action in which only the object ID from the object sensor unit is combined with the action and the user state is not included.

The user behavior record obtained as described above is used as training data, and the control data generation unit 106 generates control data. Hereinafter, an operation example of the control data generation unit 106 will be described with reference to the flowchart of FIG. First, the control data generation unit 106 selects a control pattern a (n) in the user sensor unit 120 (step S501). The control pattern is a pattern for performing power control of the user state measurement unit 121 of the user sensor unit 120, and here is a combination of on / off control of the myoelectric potential sensor. There are 2 ^h control patterns, where h is the number of user state measurement units 121. Here, two myoelectric potential sensors are used, and there are four control patterns a (1), control pattern a (2), control pattern a (3), and control pattern a (4).

  Next, the control data generation unit 106 selects one user behavior record as training data from the plurality of user behavior records stored in the user behavior storage unit 105 (step S502). As described above, the training data includes a plurality of continuous user actions, and what kind of actions these are already estimated is already known. Each user action is combined with a known action, as described above, a user state as an output from the user sensor unit and an object ID as an output from the object sensor unit.

Next, the control data generation unit 106 sequentially selects the second and subsequent user actions in the selected training data (step S503). Thereafter, the control data generation unit 106 measures the myoelectric potential sensor to be turned on corresponding to the selected control pattern a (n) from the output of the user sensor unit included in the selected user action. select a value, and output u _t users sensor portion in a state that is controlling the control pattern a (n). The control data generating unit 106, an object ID from the object sensor unit included in the selected user behavior, and the output e _t of the object sensor unit (step S504).

  The above is performed for all second and subsequent user actions in the selected training data (step S505). Subsequently, these are performed for all user action records (training data) stored in the user action storage unit 105 (step S506).

In the set control pattern, after extracting a plurality of (e _t , u _t ) from all the training data and from all the target behavior data, first of all the extracted sets of patterns by the predetermined pattern recognition method (E _t , u _t ) is used to estimate this behavior t ′ (changed user behavior) with ( _et , u _t ) newly associated with each behavior data, and the correspondence between the estimated behavior t ′ and the training data A recognition rate indicating a matching rate with the action t to be obtained is obtained for each action. The behavior t in the case where the control (control pattern a (n)) set in (e _t−1 , u _t−1 ) corresponding to the preceding behavior is estimated as the recognition rate thus obtained is estimated. The performance is set (step S507). Here, weighting may be set for each action, and this may be multiplied by the recognition rate to obtain the performance. The performance obtained in this way is a measure of how effective the action estimation obtained as a result of the control of the user sensor unit (user state measurement unit) by the set control pattern a (n) is effective. As described above, the performance based on the recognition rate obtained by pattern recognition is an amount representing how clearly the target action can be distinguished among a plurality of actions.

  The above is performed for all control patterns a (n) (step S508). As a result, as shown in part in Table 1 below, the output from each object sensor unit (object ID), the recognition rate for the control state of each sensor in the user sensor unit, the performance multiplied by the weight, and the corresponding control In this case, a table is generated with the power consumption in the user sensor unit and the corresponding action as one record. In Table 1, first, the number of myoelectric sensors that are turned on is used as the power consumption. In Table 1, since the action vvv and the action xxx are not important, the weight is set to 0.1. The action yyy has a weighting of 1, and the action zzz has a weighting of 2 as important.

  From the table obtained in this way, the control data generation unit 106 determines the optimal control pattern for each action so that the power consumption can be suppressed within a range in which the action can be estimated (recognized) more accurately. The generated control data is stored in the control data storage unit 107. In other words, in a series of user action records (training data) having T actions, the control pattern in which the total power consumption is the maximum within the range of the specified value is the optimum pattern described above. To do.

That is, the performance obtained in each action t of T actions is denoted by r _t , the power consumption in each action t is denoted by c _t, and the specified value that defines the total range of power consumption is denoted by b, After satisfying (c ₁ + c ₂ +... + C _t +... + C _T ) / T ≦ b... (1) ”,“ (r ₁ + r ₂ +... + R _t +. The combination of the control patterns that maximizes “+ r _T ) / T (2)” may be obtained.

In order to calculate this, the number of control patterns for controlling the user state measuring unit of the user sensor unit is x, the number of action patterns is T, and x ^T combinations are considered. What is necessary is just to obtain | require the combination which satisfies and maximizes Formula (2). When the number of combinations increases, a method of first obtaining a combination that maximizes Expression (2) and changing the combination so as to satisfy Expression (1) can be considered.

  Table 2 below shows an example (partial) of control data composed of combinations of control patterns obtained in this way. Here, T = 13 and b = 1.

  Using the control data prepared as described above, the user sensor control unit 108 controls the user sensor unit 120 as shown in FIG. First, for example, when the output from the object sensor unit 130A (object ID “A”) and the output from the user sensor unit 120 are received, the user behavior is estimated (step S601), and a control pattern corresponding to the estimated user behavior is set. Selection is made based on the control data (step S602). Thereafter, the user sensor unit 120 is controlled by notifying (sending) the determined control pattern to the user sensor unit 120 (step S603). In the user sensor unit 120 notified of the control pattern, the control unit 124 controls the power state (ON / OFF) of the user state measurement unit 121 based on the notified control pattern.

  According to the sensor node control system of the present embodiment that performs control as described above, the user behavior estimation result is effective in a control state in which the total power consumption of the user state measurement unit in the user sensor unit is set to a predetermined value or less. The user sensor unit (user state measurement unit) that controls the highest performance is controlled. As a result, more data effective for estimating a highly effective behavior can be obtained immediately with reduced power consumption. Therefore, in a user sensor unit (sensor node) that is driven using a battery, more effective data can be immediately communicated.

  Here, the weighting used when obtaining the performance will be briefly described. The weighting may be set according to the importance of the action. For example, if you want to estimate more reliably what the user is doing, or if you want to more reliably estimate the state of the object (life goods), increase the weight of the corresponding action. If state estimation is not very important, the weighting can be reduced.

Next, pattern recognition will be briefly described. For pattern recognition, a nearest neighbor method, a neural network, a support vector machine, or the like may be used (see Non-Patent Document 1). The case where the nearest neighbor method is used as pattern recognition will be described. First, in order to estimate the category of the vector X = (a ₁ , a ₂ ,..., A _I ) ^T belonging to a certain unknown category (action). , Calculate the inter-vector distance with a series of n vectors X ₁ , X ₂ ,..., X _N whose categories are known, and select the largest category among the categories of k vectors with the smallest distance. presume.

  Next, an operation example of the user behavior state estimation unit 101 will be briefly described. For example, when an object to which the object sensor unit 130A is attached is subjected to a predetermined operation such as being lifted, the operation state detection unit 131 of the object sensor unit 130A performs the received operation as shown in the flowchart of FIG. Detection is performed (step S701). When the operation state detection unit 131 detects an operation, the transmission / reception unit 133 transmits the object ID stored in the storage unit 132 to the sensor node control device 100 (transmission / reception unit 103) under the control of the control unit 134 (step S103). S702).

  When the object ID transmitted from the object sensor unit 130A is received by the transmission / reception unit 103 of the sensor node control device 100 (step S703), the received object ID is received under the control of the communication control unit 109 that detects this state. The received data is stored in the received data storage unit 104 together with the time information (step S704). When the communication control unit 109 stores the data in the reception data storage unit 104, the communication control unit 109 controls the transmission / reception unit 103 and transmits an instruction to transmit the measurement value to the user sensor unit 120 (step S705).

  When the transmission / reception unit 123 of the user sensor unit 120 receives an instruction to transmit the measurement value, the control unit 124 transmits the measurement value (myoelectric potential) of the user state measured by the user state measurement unit 121 from the transmission / reception unit 123 to the sensor. It transmits to the node control apparatus 100 (step S706). When the measurement value transmitted from the user sensor unit 120 is received by the transmission / reception unit 103 of the sensor node control device 100 (step S707), the received measurement value is received by the control of the communication control unit 109 that has detected this state. It memorize | stores in the memory | storage part 104 (step S708).

  When the object ID and the measurement value are stored in the reception data storage unit 104 as described above, the user behavior state estimation unit 101 stores the information in the state storage unit 102 based on the information on the measurement value and the object ID. The user's action state is estimated by referring to the stored reference data (step S709). Based on the obtained measurement value and object ID, the user's action taken out from the state storage unit 102 as corresponding to this set is the result of estimating the user state.

  Next, estimation of a user's action state using the obtained object ID, measurement value, and the like and reference data stored in the state storage unit 102 will be described in more detail. For example, the user behavior state estimation unit 101 estimates the user behavior state from the reference data by pattern recognition using the obtained object ID and measurement value as a feature vector. For this pattern recognition, a method such as a nearest neighbor method, a neural network, and a support vector machine can be used similarly to the pattern recognition described above.

Hereinafter, a case where a neural network is used as an example of pattern recognition for behavior estimation will be described. First, the category c based on the user's action state and the object ID of the object operated by the user is expressed as “c˜ = f (X, W), X = (a ₁ , a ₂ ,..., A _i , y)”. ^T , W = (w ₁ , w ₂ ,..., W _i ) ^T (1) ”. This is a case where the category c is constituted by a nonlinear function. Here, c˜ is a numerical value of the user's action state and the category of the object ID operated by the user. For example, the action “user x picks up object y” is 1, “user x etc. There act of placing the object y "and 2, f (X, W) is a non-linear function of X and W, W is the coefficient _{vector, a 1, a 2, ···} , a i is i pieces The measured value by the user sensor unit and the value obtained by the object sensor unit, y is a digitized object ID. Although one object is operated here, a plurality of objects may be used. In this case, a plurality of object IDs are provided.

  The coefficient vector W is determined using training data X, c, c˜ for a certain period. This determination will be described with reference to the flowchart of FIG. First, a vector X is constituted by the measured value (user state data) by the user sensor unit and the value (object ID) obtained by the object sensor unit (step S801).

Next, “E = ½Σ (c˜−c) ² ... (2)”, which is the sum of squares of these differences as an error between the category value c˜ estimated by the neural network and the actual category value c, is calculated. (Step S802), and using the calculated error, “dΕ / dW = Σ (c˜−c) · ∂f (X, W) / ∂W (3)” is calculated (step S803), The coefficient vector W is updated by “W = W + ΔW, ΔW = −αdΕ / dW (4)” (step S804). Α is a positive real number, and ΔW is an update amount of W.

  Next, when the wrinkle calculated in step S802 is smaller than the preset value ε, the coefficient vector W updated in step S804 is used for the category c based on the user behavior state and object ID used for state estimation. The coefficient vector W is determined. The category value c˜ may be estimated by a neural network according to the equation (1) using the coefficient vector W thus determined.

  By the way, even when an operation is performed on the same object, the data transmitted from the object sensor unit installed on the object has the same object ID, but the measurement obtained from the user sensor unit as a result of the transmission of the object ID. The value, for example myoelectric potential, reflects the state of different operations on the object. Therefore, the behavior indicating which object the user has manipulated is estimated by the user behavior state estimation unit 101 according to the present embodiment based on the measurement value (myoelectric potential) of the user state obtained from the user sensor unit. it can.

  For example, as shown in FIG. 9, a combination of a measurement result (a1, a2) of myoelectric potential 1 and myoelectric potential 2 from a user sensor and an object ID (y) from an object sensor is used as a feature vector, and a database is used. The action that “the user picked up the object y” can be estimated by the recognized pattern recognition. In addition, even if the action is directed to the same object y, “the user placed the object y” due to the difference (a3, a4) in the measurement results of myoelectric potential 1 and myoelectric potential 2 measured by the user sensor. The behavior can be estimated. Also, it can be estimated at what time these are made based on the information of the time when the object ID from the object sensor is received.

It is a block diagram which shows the structural example of the sensor node control system in embodiment of this invention. 3 is a configuration diagram illustrating a configuration example of reference data stored in a state storage unit 102. FIG. It is explanatory drawing explaining the environment where 13 object sensor part AM is arrange | positioned. It is explanatory drawing which shows an example of the user action record memorize | stored in the user action memory | storage part. It is a flowchart for demonstrating the operation example of the sensor node control system in this Embodiment. It is a flowchart for demonstrating the operation example of the sensor node control system in this Embodiment. It is explanatory drawing explaining the operation example of the user action state estimation part 101 in the sensor node control system of embodiment. It is a flowchart explaining determination of the coefficient vector W in the type | formula which shows the category c by a user's action state and object ID. It is explanatory drawing which shows the example of a state estimation.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 100 ... Sensor node control apparatus, 101 ... User action state estimation part, 102 ... State storage part, 103 ... Transmission / reception part, 104 ... Reception data storage part, 105 ... User action storage part, 106 ... Control data generation part, 107 ... Control Data storage unit 108 ... User sensor control unit 109 ... Communication control unit 110 ... Input unit 120 ... User sensor unit 121 ... User state measurement unit 122 ... Storage unit 123 ... Transmission / reception unit 130A, 130B, 130C ... object sensor part, 131 ... operation state detection part, 132 ... storage part, 133 ... transmission / reception part, 134 ... control part.

Claims (6)

An object sensor unit that is provided on the object and detects an operation received by the object and transmits an object ID for identifying itself;
A user sensor unit that includes a state measurement unit that is mounted on a user and measures the state of the user, and that transmits a user state measured by the state measurement unit;
A user action storage unit storing a plurality of user actions indicating the action state of the user corresponding to the object state transmitted from the object sensor part and the user state transmitted from the user sensor part;
It is obtained by changing the control state of the state measurement unit corresponding to the user behavior within a change range in which the power consumption of the state measurement unit is a predetermined value or less, and changing the user state corresponding to the changed control state. The control in which a plurality of changed user behaviors estimated based on a plurality of changed user states is obtained, and the changed user behavior in which the recognition rate indicating the matching rate between the obtained changed user behavior and the user behavior is maximized is estimated. A control data generating unit for generating and generating a state in the change range;
A sensor node control system comprising: at least a user sensor unit control unit that controls the user sensor unit in a control state generated by the control data generation unit. The sensor node control system according to claim 1,
The sensor data control system, wherein the control data generation unit obtains the recognition rate by pattern recognition of the changed user state with respect to a plurality of user actions stored in the user action storage unit. In the sensor node control system according to claim 1 or 2,
The control data generation unit
The sensor node control system, wherein the control state is generated by weighting the recognition rate for each corresponding user action. The object ID for identifying the object sensor unit transmitted from the object sensor unit provided on the object in order to detect the operation received by the object, and the user ID transmitted from the user sensor unit attached to the user A first step of storing a plurality of user behaviors indicating the user behavior states corresponding to the user states being measured by the state measurement unit of the user sensor unit;
It is obtained by changing the control state of the state measurement unit corresponding to the user behavior within a change range in which the power consumption of the state measurement unit is a predetermined value or less, and changing the user state corresponding to the changed control state. The control in which a plurality of changed user behaviors estimated based on a plurality of changed user states is obtained, and the changed user behavior in which the recognition rate indicating the matching rate between the obtained changed user behavior and the user behavior is maximized is estimated. A second step of determining and generating a state in the change range;
A sensor node control method comprising at least a third step of controlling the user sensor unit in the control state generated in the second step. The sensor node control method according to claim 4, wherein
In the second step, the recognition rate is obtained by pattern recognition of the changed user state with respect to a plurality of user actions stored in the user action storage unit. In the sensor node control method according to claim 1 or 2,
In the second step,
The sensor node control method, wherein the control state is generated by giving a weight to the recognition rate for each corresponding user action.

Images