JP4829081B2 - Things determining apparatus, things determining method, program, and recording medium - Google Patents

Description

  The present invention uses a general-purpose sensor node that does not fix the intended use, and uses the data such as acceleration, illuminance, and temperature acquired from the sensor node to determine the object on which the sensor node is mounted and the state of the object. The present invention relates to an apparatus, a thing determination method, a program, and a recording medium.

  The most prominent field of research on state estimation using a sensing device is ADL (Activity in Daily Living) estimation. There is a study that recognizes 20 types of user actions such as “walking”, “watching TV”, and “brushing teeth” from a waveform measured with a user carrying an acceleration sensor (see Non-Patent Document 1, for example). . In addition, research on recognizing a person's movement using an image of a person photographed by a camera from an image processing approach has also been performed (for example, see Non-Patent Document 2).

  However, since these studies estimate by learning using training data, a large amount of training data is required to realize application under various environments. Furthermore, under the environment assumed by these studies, for example, even if sensor nodes are installed on doors of various sizes, it is necessary to recognize all of them as doors. That is, since there are many variations in the object itself and the object environment, estimation using training data is difficult.

  On the other hand, there are many studies that express events in the ubiquitous computing environment by paying attention to state transition diagrams (see, for example, Non-Patent Document 3). In such a study, the state in the home is modeled using a state transition diagram. For example, in the state transition diagram showing the bathroom usage status, the state transitions between the two states of “taking a bath” and “not taking a bath” by an event such as opening / closing of a bathroom door or entry / exit of a user's bathroom. Modeled. By using state transition diagrams, it is possible to create abstract models compared to signals obtained from actual sensors such as “door opening / closing”, so it is possible to create models applicable to various environments. It is believed that there is.

L.Bao and S.S.Intille, “Activity Recognition from User-Annotated Acceleration Data”, Proc.PERVASIVE 2004, p.1-17, 2004 J. Yamato, J. Ohya, and K. Ishii, “Recognizing Human Action in Time-Sequential Images using Hidden Markov Mode 1”, Proc. IEEE Conf. Computer Vision and Pattern Recognition, p.379-385, 1992 Kaori Fujinami, Tetsuo Yamazaki, Isumi Kobayashi, Harumi Hasebe, Tatsuo Nakajima, “Real World Modeling and Application to Context Storewareness at Home”, Multimedia, Distributed, Collaboration and Mobile (DICOMO2004) Symposium, P.527-530

  However, the conventional sensor nodes to be attached to the objects are produced specifically for each object, and there is a problem that it is difficult for a person without specialized knowledge to attach the sensor node. For example, it may be possible to install a pressure sensor on a chair to determine whether a person is sitting, but a person without expert knowledge understands the pressure sensor and installs the pressure sensor in an appropriate position. It is difficult.

  The present invention has been made in order to solve the above-described problems, and uses a general-purpose sensor node that does not fix a target application, and can determine an object on which a sensor node is mounted and a state of the object. An object of the present invention is to provide a thing determination method, a program, and a recording medium.

The present invention is an object determination device for estimating an object to which a sensor node composed of a plurality of sensors is attached and a state of the object, and data for extracting a section where a change occurs from sensor data received from the sensor node A change extracting means, an event specifying means for identifying an event that is an event of an object from changes extracted by the data change extracting means, and events occurring at substantially the same time among a plurality of events obtained by the event specifying means An event set is obtained by grouping the event sets, and the event set specifying means for obtaining the event set sequence by arranging the event sets in time series, and a plurality of descriptions describing the relationship between the state of the object and the state transition that can occur in the object for each object Storage means for storing a state transition diagram in advance, and a likelihood of obtaining the event set sequence for each state transition diagram By obtaining the said sensor node is one having an object identification means for identifying the state of the object and the loaded object.
Further, in one configuration example of the thing determining apparatus according to the present invention, the state transition diagram includes information on an event that is expected to be acquired by each sensor of the sensor node when the transition between states occurs. The object identification means is configured to include all of the state transition diagrams for each event set of the event set sequence based on event information described accompanying the transition between states. A similarity calculation means for calculating a similarity with a transition between states, a transition transition selection means for selecting a transition between states that can correspond to each event set of the event set sequence, and a similarity calculation means Score calculation means for calculating a score for the event set sequence for each state transition diagram from the calculation result and the selection result of the transition transition selection means, and the calculated score is the highest. An object corresponding to a high state transition diagram is identified as an object to which the sensor node is attached, and a series of transitions between states selected by the inter-state transition selection means with respect to this state transition diagram is represented in the state of the identified object. And determining means for making a series.

The present invention also relates to an object determination method for estimating an object to which a sensor node composed of a plurality of sensors is attached and a state of the object in a computer including a CPU and a storage device. A data change extraction procedure for extracting a section where a change has occurred from sensor data, an event identification procedure for identifying an event that is an object event from the changes extracted by the data change extraction procedure, and an event identification procedure The event set is obtained by grouping together events that occur at almost the same time from a plurality of events, obtaining the event set, and arranging the event set in time series order to obtain the event set sequence, the object state and the object For each of a plurality of state transition diagrams in which the relationship between possible state transitions is described in advance for each object, The CPU is caused to execute an object identification procedure for identifying the object to which the sensor node is attached and the state of the object according to a program stored in the storage device by obtaining the likelihood that the event set sequence is obtained. It is a thing.
Further, in one configuration example of the thing determination method of the present invention, the state transition diagram includes information on events that are expected to be acquired by the sensors of the sensor node when the transition between states occurs. The object identification procedure includes all of the state transition diagrams for each event set of the event set sequence based on event information described in association with the transition between states. A similarity calculation procedure for calculating a similarity with a transition between states, a transition selection procedure between states for selecting a transition between states that can correspond to each event set of the event set sequence, and a similarity calculation procedure A score calculation procedure for calculating a score for the event set sequence for each state transition diagram from the calculation result and the selection result of the inter-state transition selection procedure, and the calculated score is the highest. An object corresponding to a high state transition diagram is identified as an object to which the sensor node is attached, and a sequence of transitions between states selected in the transition transition procedure between states for this state transition diagram is represented in the state of the identified object. And a determination procedure to be a series.

Further, the thing determination program of the present invention is a data change extraction procedure for extracting a section in which a change occurs from the sensor data received from the sensor node, and an object event from a change extracted by the data change extraction procedure. Event identification procedure that identifies a certain event and events that occur at almost the same time from the multiple events determined in this event identification procedure are grouped together to obtain an event set, and this event set is arranged in chronological order. Determine the likelihood that the event set sequence can be obtained for each of a plurality of state transition diagrams in which the relationship between the event state and the state transition that can occur in the object is described in advance for each object. Object identification procedure for identifying the object on which the sensor node is mounted and the state of the object The, in which so as to be executed by a computer.
In addition, the recording medium of the present invention is characterized in that an event determination program is recorded.

  According to the present invention, the user does not need to have knowledge about the sensor node, and only by attaching a general-purpose sensor node that does not fix the intended use to an object such as furniture, the object on which the sensor node is attached and its The state of the object can be estimated. As a result, according to the present invention, even a user who has no specialized knowledge can easily construct a base for a ubiquitous environment simply by attaching a sensor node to an object.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. The present embodiment relates to a system that estimates an object to which a sensor node is attached and what state the object is in. This is particularly relevant when there is no training data for acceleration, illuminance, and temperature signals.

  Here, the estimation of an object is to determine what the object is, and examples of the object include furniture and tools that exist in daily life such as chairs, desks, doors, and cups. Some objects have different operations depending on the part. In such a case, estimation is performed for each part where the sensor node is mounted. For example, a door knob and a door plate are greatly different in acceleration signals obtained when opened. The state of an object depends on what the object is, but it is a general state unique to the object. For example, a chair has a state where a person is sitting and a state where a person is not sitting is there. The door has an “open state” and a “closed state”.

  In the environment assumed by the present embodiment, general-purpose sensor nodes are attached to various objects in the room. The sensor node need not be given information about the object on which it is mounted. A user operates in this environment, and the sensor node collects data obtained by sensing for a certain period of time. Then, the estimation of the object is realized by comparing the metadata of various objects prepared in advance with the sensing data and determining the likely metadata from which the sensing data can be observed.

  In the present embodiment, the metadata of an object is expressed using a state transition diagram. The object state transition diagram is a graph describing the relationship between the state of the object and the state that can occur in the object. Furthermore, in this embodiment, by determining the object, it is possible to simultaneously estimate the state of the object to which the sensor node is attached.

  First, a general assumed environment of the present embodiment will be described. In the present embodiment, sensor nodes are attached to various objects in the environment. The sensor node accumulates the collected data in its own memory, or accumulates it in an external storage via wired or wireless communication. Various sensors are mounted on the sensor node. Examples of such sensors include a temperature sensor, an illuminance sensor, a pyroelectric sensor, an acceleration sensor, and a geomagnetic sensor. In this way, the sensor node is equipped with various sensors and can detect changes in various physical quantities related to the object, and thus functions as a general-purpose sensor node. That is, since the sensor node does not select the object to be mounted, the object estimation can be realized only by pasting the sensor node on the object by a user who has no specialized knowledge.

  In the present embodiment, it is assumed that a floor pressure sensor may be installed on the floor in the environment. The floor pressure sensor can observe the presence or absence of pressure on the floor with a certain degree of resolution (in this embodiment, the sensor actually used can observe the presence or absence of pressure on the floor with a resolution of 180 mm square. it can).

Next, before explaining the method of expressing the object state transition diagram, the relationship between the physical quantity change detected by the sensor node, the event that is the object event, the object state, and the transition between states will be described with reference to FIG. .
FIG. 1 is a waveform diagram showing an output obtained from a sensor node attached to a door when a series of transitions of “open”, “close”, and “open / close” are repeated twice. acc_x represents the output of the acceleration sensor in the X-axis direction, acc_z represents the output of the acceleration sensor in the Z-axis direction, and illumi represents the output of the illuminance sensor. In FIG. 1, waveforms of other sensors are omitted.

cx _i and cz _i (i = 1,..., 6) are changes in the waveforms obtained from the acceleration sensors in the X direction and the Z direction, respectively, and are simply referred to as changes hereinafter. The phenomenon identified by analyzing these changes is called an event. A method for extracting changes in sensor data and a method for identifying events by analyzing the extracted changes will be described later. For an acceleration sensor, an event is obtained by analyzing a change in triaxial acceleration, and for an illuminance sensor, an event is obtained by analyzing an illuminance change.

em _i (i = 1,..., 6) is an event that can be obtained from the acceleration sensor. For example, the events that can be obtained from the acceleration sensor include “moving horizontally” and “rotating”. ei _i (i = 1,..., 6) is an event that can be obtained from the illuminance sensor. Events that can be obtained from the illuminance sensor include “increased illuminance” and “illuminance changes”.

A group of events that occur at almost the same time on different sensors is called an event group. Here, an event that occurs at approximately the same time refers to an event of another sensor that occurred within a predetermined time range from the time when the event of this reference occurred when the event of that sensor was used as a reference. . In the example of FIG. 1, g _i (i = 1,..., 6) is an event set. For example, g ₁ is an event set including events em ₁ and ei ₁ . However, in FIG. 1, the event of the bed pressure sensor is ignored for the sake of simplicity.

The event set is a collection of events obtained by analyzing changes extracted from the sensor data. By performing these processes, g _i (i = 1,..., 6) is obtained from the sensor data. A series of event sets such as Here, the event set corresponds to a transition in the state transition diagram, and corresponds to a fixed state in the state transition diagram while the event set does not occur. For example, event sets g ₁ and g ₄ correspond to the “open” door transition. In other words, by finding such an association, it is possible to determine a state transition diagram that will obtain the change obtained by the sensor node.

In the example of FIG. 1, “opening / closing” the door is one transition. In general, it is more natural to open and close two phenomena, “open” and “close”, in succession. However, as can be seen from the waveform such as cx ₃ in FIG. 1, it is impossible to recognize from the sensor data that two transitions of “open” and “close” have occurred successively. This is because training data is not prepared in advance, but if the transition of “opening and closing” can be recognized, it can be recognized that the door has opened, so there is no problem. As described above, a phenomenon that occurs continuously without a gap is regarded as one transition.

  Next, a state transition diagram prepared in advance for object estimation will be described. When actually describing a state transition diagram of an object, the shape of the state transition diagram is determined by extracting the transition that is generally considered to occur in the object and the state of the object and connecting them. That is, the state transition diagram is created manually. At this time, it is necessary to consider the continuous transition as described above.

  FIG. 2 shows a door state transition diagram as an example. “Closed” represents a state in which the door is closed, and “Opened” represents a state in which the door is open. “Open” represents a transition between states in which the door is opened, “Close” represents a transition between states in which the door is closed, and “Open-and-shut” represents a transition between states in which the door is opened and closed. “Knock” represents a transition between states in which the door is knocked.

  A balloon B coming out of a transition “Open” represented by an arrow represents a set of events when the transition occurs in a time series, and B is a value of the transition. In the example of FIG. 2, the transition “Open” describes two event sequences. The first represents a sequence of events in which a person walks through a door while walking. Represents a sequence of events where a person walks in front of the door, stops, opens the door, and then walks away from the door.

  That is, “floor.change.walk.through” represents an event in which a person walks while walking, “acc.change.rotate” represents an event corresponding to a rotational movement when the door is opened, and “illumi” .Change "represents an event that the illuminance has changed due to the door being opened. “Floor.change.walk.here” represents an event that a person walked to the door and stopped, and “floor.change.walk.away” represents an event that the person walked away from the door. Yes. Since a plurality of cases can be described for one transition in this way, various situations can be considered.

  FIG. 3 shows a state transition diagram of the chair as another example. In the example of FIG. 3, “Sat” represents a state where a person is sitting, and “Not-sat” represents a state where a person is not sitting. “Sit” represents a transition between states that a person is sitting, and “Get-up” represents a transition between states that a person has stood up. Details of how to write transition values will be described later.

  A transition in the state transition diagram has as its value a series of events that will be acquired by each sensor when that transition occurs. The value is a series of events such as, for example, when the illuminance sensor and the acceleration sensor detect a change at the same time and then the floor pressure sensor detects a person walking. As a representation of the time relationship between events, there is a time interval logical expression as shown in FIG. 4 proposed by Allen (J.F. Allen).

  For example, as shown in FIG. 4, a relationship where there is a white rectangle indicating the time range of event Y after a black rectangle indicating the time range of event X, that is, a time relationship in which event X occurs before event Y is represented. Is described as “X before Y”. To express the relationship in which the black rectangle indicating the event X and the white rectangle indicating the event Y overlap with each other in the same length, that is, the relationship in which the events X and Y occur in the same time range, “Xequal Y ”. In order to express the relationship in which the black rectangle indicating the event X and the white rectangle indicating the event Y are connected, that is, the relationship in which the event Y occurs simultaneously with the end of the event X, “X meets Y” is described. “X overlaps Y” is used to represent a relationship in which a part of a black rectangle indicating event X and a part of a white rectangle indicating event Y overlap each other, that is, a relationship in which events X and Y overlap. Is described.

In order to express the relationship in which the black rectangle indicating the event X is placed in the center of the white rectangle indicating the event Y, that is, the relationship in which the event X occurs in the middle of the event Y, “X duration Y” Is described. To express a relationship in which a black rectangle indicating event X and a white rectangle indicating event Y are aligned with each other, ie, the time at which events X and Y occur is the same, “X “starts Y”. To represent a relationship in which the black rectangle indicating event X and the white rectangle indicating event Y are aligned and overlapped vertically, that is, the relationship that the time at which events X and Y end is the same. "Finishes Y".
In the present embodiment, the temporal relationship of event sequences that may occur at the time of transition is described as a transition value using the time interval logical expression as described above.

  Next, a procedure for identifying an object on which a node is mounted from sensor data of a certain sensor node will be described. FIG. 5 is a block diagram showing a configuration example of the thing determining apparatus according to the present embodiment, and FIG. 6 is a flowchart showing the operation of the thing determining apparatus of FIG. In FIG. 5, 1 is an object determination device, 2 is a sensor node equipped with a plurality of sensors such as a temperature sensor, an illuminance sensor, a pyroelectric sensor, and an acceleration sensor, 3 is a floor pressure sensor that measures floor pressure, and 4 is a sensor. This is a sensor data storage that receives sensor data of the node 2 and the floor pressure sensor 3 by wire communication or wireless communication and temporarily stores them.

  The thing determination apparatus 1 includes a data acquisition unit 10, a data change extraction unit 11, an event identification unit 12, an event set identification unit 13, an object identification unit 14, and a storage unit 15.

First, the data acquisition means 10 of the thing determination apparatus 1 acquires the sensor data of the sensor node 2 and the bed pressure sensor 3 from the sensor data storage 4 (step S1 in FIG. 6).
Subsequently, the data change extraction unit 11 extracts a section in which a change occurs from the sensor data acquired by the data acquisition unit 10 (step S2).

The event identification unit 12 identifies the event by analyzing the change extracted by the data change extraction unit 11 (step S3). An event is an event that can be acquired by various sensors such as “the illuminance increases”.
Then, the event set specifying unit 13 obtains an event set by grouping events that are occurring almost simultaneously from the plurality of events obtained by the event specifying unit 12, and obtains the obtained event set in chronological order. A sequence of event sets is obtained by arranging them (step S4).

  A state transition diagram is registered in the storage unit 15 for each object that can be expected to be mounted with the sensor node 2 in advance. The object identification unit 14 identifies the object by obtaining the likelihood that the sequence of event sets obtained in step S4 can be output for each state transition diagram of the object stored in the storage unit 15 (step S5). .

Next, the above operation of the thing determining apparatus 1 will be described in detail. In the thing determination apparatus 1, an event is specified by analyzing the signal obtained from the sensor. Prior to the description of the event identification, the operation of the data change extraction unit 11 (step S2) will be described. FIG. 7 is a flowchart showing the operation of the data change extraction means 11.
Sensor data obtained from various sensors always has a small vibration that becomes noise, and the amplitude of the noise often changes with time, and there are individual differences in the amplitude of the sensor. Also, the vibration is not periodic. A method for extracting changes in such sensor data will be described.

The sensor node 2 and the bed pressure sensor 3 perform sampling at regular time intervals (60 msec in the present embodiment). That is, sensor data is output from the sensor node 2 and the bed pressure sensor 3 at regular intervals.
The data change extraction means 11 of the thing determination apparatus 1 calculates the change amount at each time point of the sensor data (step S10 in FIG. 7). A method of calculating the amount of change at a certain point in time (a point in the sampling sequence obtained by the sensor node) p will be described with reference to FIGS. However, in this embodiment, the following procedure is applied to hourly data obtained from each sensor.

  Here, a window W1 having a certain width is considered. A difference between the maximum value and the minimum value of the signal intensity in the window W1 is set as a change amount ΔD of the sensor data (FIG. 8B). However, as shown in FIG. 8A, the window W1 is slid along the time direction so that the time point p is included in the window W1, and the largest change amount is set as a change amount ΔD at the time point p.

  Subsequently, the data change extraction unit 11 subtracts noise from the change amount ΔD at each time point. Here, the size of the window W1 is 15 samples in this embodiment. The noise is calculated from the amount of change around the time point p. Specifically, as shown in FIG. 8C, a large window W2 centered at the time point p is divided into a plurality of small windows W3 having the same width, and the amount of change in each window W3 is shown in FIG. (In this embodiment, the size of the window W2 is 1000 samples, and the number of the windows W3 is 10). However, in this case, the window W3 is not moved, and the difference between the maximum value and the minimum value of the signal intensity in the window W3 is set as the amount of change.

  Then, the data change extraction unit 11 sets the change amount that appears most frequently among the change amounts calculated in the plurality of windows W3 as noise, and subtracts the noise from the change amount ΔD at the time point p obtained earlier, thereby obtaining the change amount at the time point p. The amount of change ΔI excluding noise is obtained. This completes the process of step S10.

Next, the data change extraction unit 11 sets a range in which the change amount exceeds the change amount threshold as a change in the sensor data. The change amount threshold value is dynamically determined by the following procedure.
First, the data change extraction unit 11 calculates the variance of the change amount ΔI obtained in step S10 (step S11). The data change extraction unit 11 compares the calculated variance with a predetermined change variance threshold (10 ^{-11 in the} present embodiment) (step S12). If the variance is smaller than the variance variance threshold, the data change extractor 11 changes to that series. (No in step S13), the process ends.

  Further, when the variance is equal to or greater than the change amount dispersion threshold, the data change extraction unit 11 determines that there is a change in the sensor data, and determines the change amount threshold (step S14). In order to determine the change threshold value, the data change extraction unit 11 rearranges the change amounts ΔI obtained at step S10 in ascending order, and recognizes from where the change in sensor data has occurred using this list of change amounts ΔI. Decide what to do.

The data change extraction unit 11 scans the list of change amounts ΔI from the top, and pays attention to the change amount ΔI _i in which the difference D _i from the next change amount ΔI becomes larger than the change amount dispersion threshold for the first time. If the difference from the next change amount is larger than the change amount dispersion threshold and the number of change amounts regarded as the same value as the change amount ΔI _i is 20% or less of the total number of change amounts ΔI, the change amount ΔI _i Is the change amount threshold value. Here, the amount of change that is regarded as the same value as the amount of change ΔI _{i of} interest is one in which the absolute value of the difference from the amount of change ΔI _i is less than or equal to the change amount dispersion threshold. When did not comply with the above conditions is will continue to scan the list, the difference between the next variation is a change amount threshold value variation amount ΔI exceeding 200% of D _i. In this way, the process of step S14 is complete | finished.

Finally, the data change extraction unit 11 extracts a range in which the change amount ΔI exceeds the change amount threshold as a sensor data change section (step S15).
As in the present embodiment, by dynamically determining the noise and the change amount threshold value, it is possible to extract changes in the sensor data corresponding to individual differences in the sensor and temporal changes in the noise. The data change extraction unit 11 detects the change as described above for each sensor of the sensor node 2.

  Next, the operation (step S3) of the event specifying unit 12 that specifies an event from the change in sensor data extracted by the data change extracting unit 11 will be described. Specifically, the event specifying unit 12 detects an event such as a temperature rise or walking.

  FIG. 9 shows a detailed event group assumed in the present embodiment. Each tree structure in FIG. 9 indicates events that can be acquired for each type of sensor. A tree having “illumi” as a root represents an event that can be obtained by analyzing a change in the data of the illuminance sensor. First, “change”, that is, a change in illuminance is the most abstract event. Specific events corresponding to the child of the event “change” include an event “brightly” in which the illuminance has increased and an event “darkly” in which the illuminance has decreased.

  A tree having “temperature” as a root represents an event that can be obtained by analyzing a change in data of the temperature sensor, and “change”, that is, a change in temperature, is the most abstract event. Specific events corresponding to the child of the event “change” include an event “increase” in which the temperature has increased and an event “decrease” in which the temperature has decreased.

  A tree having “acc” as a root represents an event that can be obtained by analyzing a change in data of the acceleration sensor, and the most abstract event is “change”, that is, a change in acceleration. As specific events corresponding to the children of the event “change”, the event “drop” that falls, the event “slide” that moves in the horizontal direction, and the event “rise” that moves in the vertical direction fall down. There is an event “fa11” and a rotation event “rotate”.

  A tree having “floor” as a root represents an event that can be obtained by analyzing a change in the data of the bed pressure sensor, and “change”, that is, a change in the bed pressure is the most abstract event. Specific events corresponding to the child of the event “change” include an event “walk” in which a person walks and an event “point” in which some operation is performed at one place.

  Next, detection of events of the acceleration sensor and the illuminance sensor will be described. First, the change of the extracted acceleration sensor is abstracted into an event indicated by a tree having “acc” as a root in FIG. The detection of “acc. Change. Rotate” and “acc. Change. Slide” is performed by integrating the acceleration twice using the Runge-Kutta method and obtaining the locus of the sensor node 2. That is, if the locus of the sensor node 2 is an arc and the central axis of the arc is substantially vertical, “acc. Change. Rotate” is assumed, and if the locus is substantially straight and horizontal, “acc. Change. Slide” is assumed.

  Detection of “acc.change.drop” and “acc.change.fall” is performed by obtaining the trajectory of the sensor node 2 and the magnitude of acceleration in the vertical direction. That is, if the trajectory of the sensor node 2 is substantially straight and vertically downward, “acc. Change. Drop” is assumed. If the trajectory is an arc and the center axis of the arc is substantially horizontal, “acc. Change. Fall” is assumed. Yes. The detection of “acc.change.rise” is performed by determining whether the direction of the locus of the sensor node 2 is vertically upward. An event that does not correspond to the above event is assumed to be “acc. Change”.

  Similarly, the extracted change in the illuminance sensor is also abstracted into an event shown in a tree rooted in “illumi” in FIG. 9. The illuminance sensor used in this embodiment outputs a value from 0 to 1024 as illuminance. In the present embodiment, when the illuminance increases by 20 or more before and after the change, “illumi.change.brightly” is set, and when the illuminance decreases by 20 or more, “illumi.change.darkly” is set. The other changes were “illumi.change”.

  Next, event detection of the bed pressure sensor 3 will be described. The bed pressure sensor 3 used in the present embodiment outputs a change in pressure on / off. The resolution is 18 × 18 cm. The event identification unit 12 clusters the ON changes obtained by the bed pressure sensor 3 in consideration of their temporal and spatial distances, and then detects a walking event from the variance of the change positions. In the clustering, the interval between the occurrence times of any two changes is within a predetermined time threshold (2 seconds in the present embodiment), and the distance between the two changes is a predetermined change distance threshold (100 cm in the present embodiment). This is done by making the items in the same group.

  For event identification, the variance of the position of the pressure change within the same group, the maximum distance between any two points within the same group, and the time difference between the first and last change within the same group. Use. Here, as the variance of the position, the average of the variance of the X coordinate and the Y coordinate of the change in the bed pressure sensor is used. The event specifying means 12 is such that the variance of the position of the change within the same group is larger than a predetermined variance threshold (2 in the present embodiment), and the maximum distance between the changes of any two points in the same group is a predetermined maximum. When it is larger than the distance threshold (1 m in the present embodiment), the walking event is “floor.change.walk”.

  In addition, the event specifying unit 12 is configured such that the variance of the position of the change in the same group is smaller than the variance threshold, the maximum distance between any two changes in the same group is smaller than the maximum distance threshold, and the first in the same group When the time difference between the last change and the last change is smaller than a predetermined change time difference threshold (2 seconds in the present embodiment), the event “floor.change.point” grounded to one point on the floor is assumed. Otherwise, it is interpreted as an event “floor.change” that some kind of change has occurred on the floor. In this way, the event specifying unit 12 can specify the event of the floor pressure sensor 3.

Next, the operation (step S4) of the event set specifying unit 13 for obtaining an event set by grouping events that occurred at the same time from the event groups of various sensors obtained by the event specifying unit 12 will be described.
The event set specifying means 13 obtains an event set from the logical sum of the event ranges that overlap in time as shown in FIG. However, even if they do not overlap, they may be included in the same event set as long as the time difference between the event occurrence times is within a predetermined event time difference threshold (0.5 seconds in the present embodiment).

  Furthermore, the event set specifying means 13 represents the temporal relationship between events included in the same event set in the time interval logical expression shown in FIG. Then, the event set specifying unit 13 creates a sequence of event sets by arranging the obtained event sets in time series.

  Next, the operation (step S5) of the object identification unit 14 for identifying an object will be described. The object identification unit 14 calculates the likelihood (score) that the series of event sets obtained from the sensor node 2 and the bed pressure sensor 3 is output for each state transition diagram of the object stored in the storage unit 15 in advance. The sensor node 2 is assumed to be attached to the object corresponding to the state transition diagram having the highest score.

  First, a general flow of a method of calculating a state transition diagram score for an event set sequence will be described. FIG. 10 is a flowchart showing the operation of the object identification unit 14. The object identification unit 14 calculates the similarity of all transitions in the state transition diagram for each event set (step S20 in FIG. 10), and considers conditions such as the shape of the state transition diagram for each event set. (Step S21), a score for the event set sequence is calculated for each state transition diagram (step S22), and the object and the state of the object are identified based on the calculated score (step S23). ). Hereinafter, the details of the processing flow of FIG. 10 will be described.

  The transition event described in the state transition diagram is represented by the time interval logical expression as shown in FIG. 4 as the transition value. On the other hand, the object identifying unit 14 obtains the similarity with the event set obtained by the event set specifying unit 13 (step S20). For example, when the time interval logical expression described in the balloon associated with the transition of the state transition diagram indicates “X meets Y” as shown in FIG. 11A, the event X and the event included in the event set When Y has a temporal relationship as shown in FIG. 11B, the error between the transition of the state transition diagram and the event set is the time (seconds) for shifting the event X or Y to satisfy “X meets Y”. It becomes. In the example of FIG. 11A and FIG. 11B, it is the difference between the end time of event X and the start time of event Y.

  In addition to such temporal similarity, the similarity score between the transition described in the state transition diagram and the event set is calculated by considering the detailed event similarity. The event similarity is calculated using the tree structure shown in FIG. For example, if the event of transition described in the state transition diagram is “acc.change.slide” and the event of the event set is “acc.change”, the similarity of these events is a tree structure. 1 is the distance between the nodes. This is because, in the example of FIG. 9, the distance from the event “slide” to the event “change” is 1 in the tree structure having “acc” as a root.

  Also, if the event of transition described in the state transition diagram is “acc.change.slide” and the event of the event set is “acc.change.drop”, the similarity of these events is 2. The reason is that, in the tree structure rooted in “acc”, in order to reach the event “drop” from the event “slide”, it is necessary to go through the event “change”, so the distance becomes 2. .

  However, in the state transition diagram, a transition event in the state transition diagram is more abstract, such as when the event of a certain transition is “acc.change” and the event of the event set is “acc.change.slide”. When the degree is large, the degree of similarity is set to zero. The final similarity between the event set and the transition event of the state transition diagram is the sum of the temporal similarity and the similarity calculated using the tree structure of FIG. Needless to say, the smaller the similarity is, the more similar the events are. The object identification unit 14 calculates the similarity between each event set and all transitions in each state transition diagram as described above.

  Next, the object identification means 14 obtains all combinations of each event set and transition mapping (step S21). For example, when there are 10 event sets arranged in time series, a series of transition combinations that can correspond to each event set is obtained. In this example, it is necessary to obtain as many as ten powers of the number of transition values included in the state transition diagram. Therefore, branch hunting is performed using the following three methods (a) to (c).

(A) The object identification unit 14 ignores inconsistent sequences in consideration of the state transition diagram. Here, the contradiction means that, for example, in the state transition diagram of the door, the opening / closing of the door, that is, the transition “Open-and-shut” occurs immediately after the transition “Open” occurs.

(B) When the object identification unit 14 determines the transition t _n of the state transition diagram corresponding to a certain event set g _i , if there are a plurality of transitions that transition to a certain state in the state transition diagram, step S20 of those transitions Select only the one with the maximum similarity calculated in.

(C) The object identification unit 14 associates event sets having similar data waveforms of the acceleration sensor and the illuminance sensor with the same transition in the event set series. For example, since the event sets g ₁ and g ₄ in FIG. 1 have similar data waveforms of the acceleration sensor and the illuminance sensor, if the event set g ₁ corresponds to a transition “Open” in a certain series, the event set g ₄ also corresponds to the same “Open” transition.

  The similarity of the waveform of sensor data is calculated using DP (Dynamic Programming) matching. Specifically, the two waveforms are normalized (the waveforms are expanded and contracted so that the maximum value of each waveform is 1 and the minimum value is 0), and then the distance between the two waveforms is obtained using DP matching. If the distance is smaller than a predetermined threshold (in this embodiment, 0.04 for the illuminance sensor and 0.07 for the acceleration sensor), the waveform is similar. As for DP matching, for example, the document “H. Sakoe, S. Chiba,“ Dynamic programming algorithm optimization for spoken word recognitioh ”, IEEE Trans. Acoust., Speech, and Signal Process., Vol. ASSP-26, p. 43-49, 1978 ".

  Then, the object identification unit 14 uses the sum obtained by adding the similarities calculated in step S20 for the transitions selected for each event set in step S21 as the score of the state transition diagram for the event set sequence (step S21). S22). The object identification unit 14 performs such score calculation for each state transition diagram of an object prepared in advance in the storage unit 15.

  Finally, the object identification unit 14 identifies the object corresponding to the state transition diagram having the highest score calculated in step S22 as the object to which the sensor node 2 is attached, and the transition selected in step S21 for this state transition diagram. Is a transition between states of the object that occurred during the data collection period (step S23).

  As described above, in the present embodiment, since the object to which the sensor node 2 is attached and the state of the object can be estimated, even a user without specialized knowledge can use a general-purpose sensor node such as furniture. By simply attaching it to an object, you can easily build a foundation for a ubiquitous environment.

  In addition, the thing determination apparatus 1 of this Embodiment is realizable with the computer provided with CPU, a memory | storage device, and the interface with the exterior, and the program which controls these hardware resources. In such a computer, a program for realizing the object determination method of the present invention is provided in a state of being recorded on a recording medium such as a flexible disk, a CD-ROM, a DVD-ROM, or a memory card. The CPU writes the program read from the recording medium into the storage device, and executes the above-described processing according to the program.

  The present invention can be applied to an object in which a sensor node is mounted and a technique for estimating the state of the object using a general-purpose sensor node.

It is a wave form diagram which shows an example of the output obtained from the sensor node in embodiment of this invention. It is a figure which shows one example of the state transition diagram of a door. It is a figure which shows one example of the state transition diagram of a chair. It is a figure which shows an example of the time interval logical expression showing the time relationship between events in embodiment of this invention. It is a block diagram which shows one structural example of the thing determination apparatus which concerns on embodiment of this invention. It is a flowchart which shows operation | movement of the thing determination apparatus which concerns on embodiment of this invention. It is a flowchart which shows operation | movement of the data change extraction means of the thing determination apparatus which concerns on embodiment of this invention. It is a figure for demonstrating the calculation method of the variation | change_quantity of the sensor data in embodiment of this invention. It is a figure which shows an example of the event in embodiment of this invention. It is a flowchart which shows operation | movement of the object identification means of the thing determination apparatus which concerns on embodiment of this invention. It is a figure for demonstrating the calculation method of the similarity degree of an event set and the transition of a state transition diagram in embodiment of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Things determination apparatus, 2 ... Sensor node, 3 ... Bed pressure sensor, 4 ... Storage for sensor data, 10 ... Data acquisition means, 11 ... Data change extraction means, 12 ... Event specification means, 13 ... Event set specification means, 14 ... object identification means, 15 ... storage means.

Claims (7)

An object determination device for estimating an object to which a sensor node composed of a plurality of sensors is attached and a state of the object,
Data change extracting means for extracting a section where a change occurs from the sensor data received from the sensor node;
From the change extracted by the data change extracting means, an event specifying means for specifying an event that is an event of an object;
Event set identification means for obtaining an event set sequence by grouping events occurring at substantially the same time from a plurality of events obtained by the event specification means, and arranging the event sets in time series,
Storage means for storing in advance a plurality of state transition diagrams describing, for each object, the relationship between the state of the object and the state transitions that can occur in the object;
An object determination device comprising: an object to which the sensor node is attached and an object identification unit for identifying the state of the object by obtaining the likelihood that the event set sequence is obtained for each state transition diagram . In the thing determination apparatus of Claim 1,
In the state transition diagram, event information that is expected to be acquired by each sensor of the sensor node when a transition between states occurs is described accompanying each transition between states,
The object identification means includes
Based on event information described accompanying the transition between states, similarity calculation means for calculating the similarity between all transitions of the state transition diagram for each event set of the event set series,
An inter-state transition selecting means for selecting an inter-state transition that can correspond to each event set of the event set series;
Score calculation means for calculating a score for the event set sequence for each state transition diagram from the calculation result of the similarity calculation means and the selection result of the transition transition selection means between states,
The object corresponding to the state transition diagram having the highest calculated score is identified as the object to which the sensor node is attached, and the state transition sequence selected by the inter-state transition selection means for the state transition diagram is And a determination means for determining a sequence of states of the identified object. An object determination method for estimating an object to which a sensor node composed of a plurality of sensors is attached and a state of the object in a computer including a CPU and a storage device,
A data change extraction procedure for extracting a section where a change occurs from the sensor data received from the sensor node;
An event identification procedure for identifying an event that is an event of an object from the changes extracted in this data change extraction procedure;
An event set identification procedure for obtaining an event set sequence by grouping events occurring at substantially the same time from a plurality of events obtained by this event specification procedure, and arranging the event set in time series,
The sensor node is mounted by determining the likelihood that the event set sequence is obtained for each of a plurality of state transition diagrams in which the relationship between the state of the object and the state transition that can occur in the object is described in advance for each object. An object determination method that causes the CPU to execute an object identification procedure for identifying an object and a state of the object according to a program stored in the storage device. In the method of determining a thing according to claim 3,
In the state transition diagram, event information that is expected to be acquired by each sensor of the sensor node when a transition between states occurs is described accompanying each transition between states,
The object identification procedure includes:
Based on the event information described accompanying the transition between the states, a similarity calculation procedure for calculating the similarity between all transitions in the state transition diagram for each event set of the event set sequence,
An inter-state transition selection procedure for selecting an inter-state transition that can correspond to each event set of the event set sequence;
From the calculation result of the similarity calculation procedure and the selection result of the inter-state transition selection procedure, a score calculation procedure for calculating a score for the event set sequence for each state transition diagram;
The object corresponding to the state transition diagram having the highest calculated score is identified as an object to which the sensor node is attached, and the sequence of transitions between states selected in the transition transition selection procedure between states for the state transition diagram is A determination procedure for determining a sequence of states of the identified object. An object determination program for operating a computer as an object determination device for estimating an object to which a sensor node composed of a plurality of sensors is attached and a state of the object,
A data change extraction procedure for extracting a section where a change occurs from the sensor data received from the sensor node;
An event identification procedure for identifying an event that is an event of an object from the changes extracted in this data change extraction procedure;
An event set identification procedure for obtaining an event set sequence by grouping events occurring at substantially the same time from a plurality of events obtained by this event specification procedure, and arranging the event set in time series,
The sensor node is mounted by determining the likelihood that the event set sequence is obtained for each of a plurality of state transition diagrams in which the relationship between the state of the object and the state transition that can occur in the object is described in advance for each object. An object determination program that causes a computer to execute an object identification procedure for identifying an object and a state of the object. In the thing determination program of Claim 5,
In the state transition diagram, event information that is expected to be acquired by each sensor of the sensor node when a transition between states occurs is described accompanying each transition between states,
The object identification procedure includes:
Based on the event information described accompanying the transition between the states, a similarity calculation procedure for calculating the similarity between all transitions in the state transition diagram for each event set of the event set sequence,
An inter-state transition selection procedure for selecting an inter-state transition that can correspond to each event set of the event set sequence;
From the calculation result of the similarity calculation procedure and the selection result of the inter-state transition selection procedure, a score calculation procedure for calculating a score for the event set sequence for each state transition diagram;
The object corresponding to the state transition diagram having the highest calculated score is identified as an object to which the sensor node is attached, and the sequence of transitions between states selected in the transition transition selection procedure between states for the state transition diagram is And a determination procedure for determining a sequence of states of the identified object.   A recording medium on which the thing determination program according to claim 5 or 6 is recorded.

Images