JP2013149104A - Action estimation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an action estimation device for allocating a high order action to an estimated basic operation.SOLUTION: An action estimation device includes: a change information acquisition part; a first estimation part; a storage part; a second estimation part; and a division determination part. The change information acquisition part acquires change information which changes in accordance with the action of an actor. The first estimation part estimates a plurality of time-sequentially arranged basic operations of the actor. The storage part stores a task flow in which k high order actions constituted of the combination of basic operations are included in a predetermined order. The second estimation part estimates the high order action constituted of the combination of the basic operations estimated by the first estimation part as a temporary high order action for every basic operation on the basis of the probability distribution in the high order action of the plurality of basic operations. The division determination part divides a plurality of time-sequentially arranged temporary high order actions into k time-sequentially arranged groups, and allocates the high order actions included in the task flow to each of those groups in accordance with the order.

Description

  Embodiments described herein relate generally to a behavior estimation apparatus.

  2. Description of the Related Art Conventionally, there is known an information processing apparatus that identifies a user's movement status based on information from an acceleration sensor and performs appropriate processing according to the identified movement status.

JP 2010-146223 A

  However, even if it is possible to identify basic motions such as “walking” and “sitting” from various sensing information represented by position and acceleration, etc. It was difficult to estimate and assign complex upper-level actions.

  The behavior estimation device according to the embodiment includes a change information acquisition unit, a first estimation unit, a storage unit, a second estimation unit, and a division determination unit. The change information acquisition unit acquires change information that changes according to the behavior of the actor. The first estimation unit estimates a plurality of basic actions arranged in a time series of the actor based on the change information. The storage unit stores a business flow including k pieces (k is an integer of 2 or more) in a predetermined order. The second estimator is configured to determine a higher-order action configured by combining the basic actions estimated by the first estimator based on a probability distribution of higher-order actions of a plurality of basic actions included in a predetermined period. Estimated as provisional higher level action for each basic action. The division determination unit divides a plurality of provisional higher-order actions arranged in a time series into k groups arranged in a time series based on the business flow, and assigns the upper actions included in the business flow in accordance with the order of each group. assign.

The block diagram of the action estimation apparatus concerning embodiment. The functional block diagram of the action estimation apparatus concerning embodiment. The conceptual diagram which shows the structure of an acceleration database and a position database. The chart which illustrates an estimation rule. The chart which shows basic operation series data w of each actor. The chart which shows the specific example which a 2nd estimation part estimates provisional high-order action. The graph which shows the probability distribution in the high-order action of basic operation. The chart which illustrates the calculation process of average information amount. The chart which illustrates the relationship between a division | segmentation point and average information content. The chart which shows the process example which changes a division | segmentation point according to a similarity. The chart which shows the example of a process which specifies the 1st best division | segmentation point. The chart which shows the process example which specifies the 2nd best division | segmentation point recursively. The example of a display which displays the high-order action to which the output part was provided with the high-order action name. The flowchart which shows operation | movement of a portable terminal. The flowchart which shows operation | movement of the management server in the case of estimating basic operation | movement. The flowchart which shows the operation | movement which a management server displays a business condition using basic operation data.

  Embodiments of a behavior estimation device will be described in detail below with reference to the accompanying drawings. FIG. 1 is a configuration diagram illustrating the configuration of the behavior estimation apparatus 1 according to the embodiment. As shown in FIG. 1, the behavior estimation apparatus 1 is configured by connecting a plurality of mobile terminals 2 to a management server 4 via a network 3, for example. The behavior estimation apparatus 1 according to the embodiment will be described by taking nursing work (medical work) as an example, but is not limited thereto, and may be applied to other work such as machine maintenance work and moving work. Good.

  The mobile terminal 2 is a mobile terminal device having a function as a computer such as a smartphone (or a mobile phone). The portable terminal 2 is attached to the waist of each of the actors such as nurses and doctors engaged in medical work, for example. The network 3 is a wireless LAN (Local Area Network) having a plurality of access points 30 for which mutual connectivity is guaranteed by, for example, Wi-Fi (registered trademark). For example, one access point 30 is arranged in each of the first ward, the second ward, and the like. The management server 4 is, for example, a PC (Personal Computer) or the like, and has a function to be described later with reference to FIG. The network 3 may have a configuration in which all devices perform wireless communication with each other, or a configuration in which some devices perform wired communication.

  FIG. 2 is a functional block diagram illustrating an outline of functions of the behavior estimation apparatus 1 according to the embodiment. The mobile terminal 2 includes a control unit 20, an operation unit 22, a detection unit 24, a communication unit 26, and a storage unit 28.

  The control unit 20 controls each unit constituting the mobile terminal 2. The operation unit 22 is, for example, a touch panel, and receives a user input to the mobile terminal 2 and outputs a processing result by the mobile terminal 2.

  The detection unit 24 is, for example, a triaxial acceleration sensor, and is a vertical direction (y-axis direction), an actor's traveling direction (z-axis direction), and a horizontal direction (x-axis direction) substantially orthogonal to the actor's traveling direction. ) Acceleration (acceleration in three directions). That is, the detection unit 24 detects an acceleration that changes according to some action (behavior) of each actor wearing the mobile terminal 2 and outputs the detected acceleration to the control unit 20.

  The communication unit 26 has a wireless communication function whose mutual connectivity is guaranteed by, for example, Wi-Fi (registered trademark), and specifies and specifies the closest access point 30 at a predetermined time interval. Information indicating the access point 30 is output to the control unit 20.

  The storage unit 28 includes a ROM (Read Only Memory) and a RAM (Random Access Memory) not shown, and stores programs executed by the mobile terminal 2, data used when the mobile terminal 2 executes the programs, and the like. . For example, the storage unit 28 stores acceleration data and access point information.

  The acceleration data is data in which the control unit 20 associates the detection time with the acceleration detected by the detection unit 24, and is one piece of change information that changes with time according to the behavior of the actor (see FIG. 3). ). The access point information (position data) is data in which the control unit 20 associates the position and time with each portable terminal 2, and is one piece of change information that changes with time according to the behavior of the actor (FIG. 3). reference). The position of each mobile terminal 2 (actor) is indicated by the position of the access point 30 located closest to the mobile terminal 2, for example.

  Each portable terminal 2 manages information such as a MAC address (Media Access Control address) that identifies the portable terminal 2 by the communication unit 26, acceleration data, and position data at predetermined time intervals (for example, every 10 mS). It transmits to the server 4.

  The management server 4 includes a communication unit 40, a storage unit 5, a processing unit 6, and an output unit 42. The communication unit 40 has a wireless communication function in which mutual connectivity is guaranteed by, for example, Wi-Fi (registered trademark). The communication unit 40 individually receives acceleration data, position data, and the like of each of a plurality of nurses and doctors transmitted from each of the plurality of mobile terminals 2 via the network 3 and outputs them to the processing unit 6. That is, the management server 4 individually acquires change information from each mobile terminal 2 by the communication unit 40. The output unit 42 is a display device such as a display, and outputs the processing result of the processing unit 6 and the like.

  The storage unit 5 includes an acceleration database (acceleration DB) 51, a position database (position DB) 52, an estimation rule 53, a basic motion database (basic motion DB) 54, a frequency distribution database (described later) using FIGS. Frequency distribution DB) 55, provisional upper action database (provisional upper action DB) 56, division time database (division time DB) 57, business flow database (business flow DB) 58, business situation database (business situation DB) 59, and the like are stored. To do. And the memory | storage part 5 memorize | stores the result etc. which the process part 6 processed, while outputting the data memorize | stored previously for performing processes, such as action estimation mentioned later, to the process part 6. FIG. The storage unit 5 can be configured by any commonly used storage medium such as an HDD (Hard Disk Drive), an optical disk, a memory card, and a RAM (Random Access Memory). Each DB may be stored in a physically different storage medium.

  The processing unit 6 includes a first management unit 61, a first estimation unit 62, a second management unit 63, a second estimation unit 64, a third management unit 65, a division determination unit 66, a name assignment unit 67, and a fourth management unit 68. And a sudden event detection unit 69. Then, the processing unit 6 receives the acceleration data and the position data received by the communication unit 40, performs the following processing using the data stored in the storage unit 5, and performs the following processing on the storage unit 5 and the output unit 42: Output.

  The processing performed by the processing unit 6 includes processing for creating the acceleration DB 51 and the position DB 52, processing for estimating the basic motion (and the basic motion sequence data w) from the acceleration DB 51 and the position DB 52, and provisional upper behavior sequence data from the basic motion sequence data w. Processing for estimating Z, processing for dividing provisional upper action sequence data Z into the number of upper actions, processing for assigning and managing names of upper actions to each divided provisional upper action sequence data Z, and provisional upper action sequence data Z is roughly divided into six processes for detecting sudden events from Z.

  First, processing in which the processing unit 6 creates the acceleration DB 51 and the position DB 52 will be described. The first management unit 61 receives, for example, acceleration data and position data received by the communication unit 40 every 10 mS. The first management unit 61 performs preprocessing such as noise processing for removing noise data and calculation of a median value within a predetermined time from acceleration data and position data. The first management unit 61 stores and manages the preprocessed acceleration data and position data in the acceleration DB 51 and the position DB 52 of the storage unit 5, respectively. The first management unit 61 may not perform noise processing and preprocessing. Further, the first management unit 61 may perform processing such as average value calculation, frequency calculation, or Fourier transform on the acceleration data.

  FIG. 3 is a conceptual diagram showing the configuration of the acceleration DB 51 and the position DB 52. As shown in FIG. 3A, in the acceleration DB 51, the passage of time and the changing acceleration are associated with each actor.

  Further, as shown in FIG. 3B, the position DB 52 includes a time when the position (access point 30) of each portable terminal 2 (actor) is specified, and a unique ID name corresponding to each access point 30 (for example, This is a database in which the ward name) and the MAC addresses of the mobile terminals 2 are associated with each other. That is, the position DB 52 indicates the time when the mobile terminal 2 (actor) specified by the MAC address is near the access point 30 specified by the ID name, in time series.

  Next, processing in which the processing unit 6 estimates basic motion (and basic motion sequence data w) from the acceleration DB 51 and position DB 52 will be described. The first estimation unit 62 receives the acceleration DB 51 and the position DB 52 of a predetermined time determined in advance via the first management unit 61, reads the estimation rule 53 (FIG. 4) stored in advance in the storage unit 5, The basic motion of each actor is estimated and output to the second manager 63. Here, the basic movement refers to basic movements (simple movements) such as “walking”, “running”, “standing position”, “sitting position”, and “recumbent position”.

  FIG. 4 is a chart illustrating the estimation rule 53. The estimation rule 53 is a determination rule for estimating a basic action of an actor using acceleration data and position data. For example, “AP (t) = 1 & Ave (AccY (t)) ≦ 100 & Ave (| AccY (t) −AccY (t−1) |)> 50 → walking” in the first line of FIG. The average value of acceleration in the y-axis direction is 100 or less, and the average value (absolute value) of the change in acceleration in the y-axis direction at time t and time t-1 is 50 It represents a rule that determines that “if it is large, it is walking”. The estimation rule 53 may be a rule created in advance by the designer as illustrated in FIG. 4 or a rule created using a machine learning technique such as a neural network.

  Then, the first estimation unit 62 applies the estimation rule 53 to acceleration data and position data for a predetermined period of time, and as shown in FIG. 5, the basic operations arranged in time series for each actor. (Basic motion sequence data w) is generated and output to the second manager 63.

  The second manager 63 stores the basic motion sequence data w received from the first estimator 62 in the basic motion DB 54 for management.

  Next, a process in which the processing unit 6 estimates the provisional higher-order action series data Z from the basic action series data w will be described. The second estimation unit 64 accepts the basic motion sequence data w via the second management unit 63, estimates the provisional higher-order action sequence data Z in which the provisional higher-order actions are arranged in time series using the frequency distribution DB 55, 3 Output to the management unit 65. The provisional superior action is a provisional superior action that may not coincide with the superior action to which the action estimation apparatus 1 finally assigns a name, and will be described later with reference to FIGS. 6 to 12. To do.

  Here, the second estimation unit 64 first reads the basic motion sequence data w for a predetermined period from the basic motion DB 54 via the second management unit 63.

  Then, as illustrated in FIG. 6, the second estimation unit 64 estimates the provisional higher action sequence data Z with respect to the basic motion sequence data w. Details of FIG. 6 will be described later. Specifically, the second estimation unit 64 reads the frequency distribution of the basic motion (probability distribution in the higher-level behavior of the basic motion: see FIG. 7) stored in advance in the frequency distribution DB 55 and reads the frequency distribution of the basic motion read out. The topic model is applied based on the above, and the provisional upper action sequence data Z is generated.

In applying the topic model, the second estimation unit 64 first divides the basic motion sequence data w = {w _i } into D by a predetermined fixed time window, and motion data d = d ₁ , d _2. ,..., D _D is created. Then, the second estimation unit 64 uses the motion data d to determine the probability that the higher level action assigned to a specific basic motion w _i is j (Z _i = j) based on the conditional probability of the following equation 1. calculate.

Here, C ^WT represents a frequency matrix (W × T) of basic motion and upper action, and C ^DT represents a frequency matrix (D × T) of motion data and upper action. W, D, and T represent the number of basic actions, the number of action data, and the number of upper actions, respectively. α and β represent parameters set by the designer.

The second estimation unit 64 assigns a higher-level action j having the maximum conditional probability value calculated by Expression 1 to the basic action w _i . That is, the second estimation unit 64 uniquely determines a higher level action for each basic action. However, the higher level action determined by the second estimating unit 64 for each basic action is shown as a provisional higher level action (label) that temporarily indicates the higher level action. May not be consistent with the top actions that give names.

  Note that the second estimation unit 64 may use an approximate calculation method such as variational Bayes or Gibbs sampling for the calculation according to Equation 1 and the assignment of a higher level action to the basic action. Hereinafter, the higher level action determined by the second estimation unit 64 is referred to as a provisional higher level action. In addition, the provisional higher level action may be simply referred to as a label.

When the provisional upper action j is uniquely determined for the basic action, the provisional upper action j is expressed as a probability distribution P (w, z) = φ ^j of the basic action w _i as shown in the following equation 2. The

FIG. 6 is a chart illustrating a specific example in which the second estimation unit 64 estimates (determines) the provisional higher level action. As shown in FIG. 6, the basic motion is assumed to be any element of a basic motion set having five types of elements, for example, {still, walking, running, bending, and elevator}. For example, the second estimation unit 64 reads the basic motion series data w (w _{1 to} w ₁₆ ) of Nurse Sato for a predetermined period from the basic motion DB 54 via the second management unit 63.

Then, the second estimation unit 64 divides the basic motion sequence data w for each predetermined fixed time window to create four motion data d = d ₁ , d ₂ , d ₃ , d ₄ , By applying the topic model to the motion data that has been processed, the provisional upper action sequence data Z is generated.

In the provisional upper action sequence data Z (label sequence data Z), one label is assigned to one basic operation. That is, the basic motion sequence data w and the label sequence data Z correspond to each other with the basic motion (or label) as a unit. For example, the label Z ₁ is assigned to the basic operation w ₁ (stationary). Therefore, in the label series data Z, for example, any of six types of labels Z ₁ ,..., Z ₆ is assigned for each basic operation, and the labels are arranged in time series.

Each label (Z ₁ ,..., Z ₆ ) is expressed by two probability distributions P (w, z) = φ ^j as illustrated by a bar graph in FIG. For example, Z ₁ which is one of the generated labels is stationary (0.5), walking (0.25), bending position (0.1), running (0.1), elevator (0.05) in this order. It is expressed as a label that has a high probability of appearance of a basic action. The probability distribution P (w, z) of each label is stored in advance in the frequency distribution DB 55 described above.

  The third management unit 65 stores and manages the label series data Z received from the second estimation unit 64 in the provisional higher-level action DB 56.

  Next, processing in which the processing unit 6 divides the label series data Z into the number of higher actions will be described. The division determination unit 66 includes a division unit 660, an average information amount calculation unit 662, a similarity calculation unit 664, an evaluation unit 666, and a determination unit 668.

  First, the division determination unit 66 reads business flow information from the business flow DB 58. The business flow DB 58 is a database that stores business flow information in advance. The business flow information includes, for example, a predetermined work name (higher action name), work order (higher action order) of nursing work (final higher action) such as “send-off”, “decubitus round”, and “operate-feed” And information including the number of tasks (the number of upper actions). In other words, the business flow DB 58 stores a business flow including k pieces (k is an integer of 2 or more) in a predetermined order of higher-order actions configured by combinations of basic operations. Note that the business flow DB 58 may store action order information with a predetermined order instead of the business flow information. For example, the action order information is information that can extract an action name, an action order, and the number of actions.

Then, the division determination unit 66 receives the label series data Z from the provisional higher-level action DB 56 via the third management unit 65, and first evaluates the label series data Z (= X ₀ ) that has not been divided.

For example, as illustrated in FIG. 8, the division determination unit 66 displays the business flow information including the upper action names, the upper action order, and the upper action number (= 3) of “send sending”, “decubitus round” and “operating sending”. read out. Then, the division determination unit 66 assigns “Send” to the first label Z ₂ according to the business flow information for the label series data Z received from the provisional higher-level action DB 56, and the label Z ₁ that follows Z _2. The average information amount calculation unit 662 calculates the average information amount when “Occult rounds” is assigned to “Z” and “operating” is assigned to Z ₃ , which is a label following Z ₁ .

Here, the average information amount calculation unit 662 calculates the average information amount I _X0 of the label series data X ₀ that has not been divided by the following Equation 3.

In Equation 3, K is the number of label types. S is the total number of labels for which the average information amount is to be calculated. freq (Z _j , S) represents the appearance frequency of label Z _j in S.

In the example shown in FIG. 8, the label series data X ₀ that is not divided includes K = 3, S = 10, freq (Z ₁ , 10) = 5, freq (Z ₂ , 10) = 2, freq (Z ₃ , 10) = 3. Then, the division determination unit 66 obtains the average information amount I _X0 = 1.485 of the label series data X ₀ that has not been divided, as shown in the following equation 4. The calculation result of the average information amount is appropriately rounded.

  Next, the division determination unit 66 sequentially sets division points Xn for dividing the label series data Z in units of labels, and reduces the average information amount of the label series data Z (= Xn) divided at the division points Xn. Calculated according to Equation 5.

In Equation 5, n indicates that the point after the nth label in the time series is the division point Xn. K is the number of types of labels for which the average information amount is to be calculated. M is the number of label series data Z after being divided. Sm is the mth group in the time series after being divided. | Sm | is the total number of labels for which the average information amount of Sm is to be calculated. freq (Z _j , Sm) represents the appearance frequency of label Z _j in Sm.

  For example, when the label sequence data Z is divided into two at one division point Xn (M = 2), the average information amount of the divided label sequence data Xn is the average information amount before and after the division point Xn. Become sum. In the case of division at two division points, M = 3.

In the example illustrated in FIG. 9, the average information amount calculation unit 662 sets the earliest (after the first label) in the order of time series with respect to the label series data Z as the division point X ₁ (division X ₁ ). set, it calculates the average information amount I _X1 of label series data X _1.

In FIG. 9, since the label series data Z is divided into two, M = 2. In this case, since the label series data (first group: divided data) before the dividing point X ₁ is composed of only one label Z ₂ , | S ₁ | = 1, freq (Z ₂ , S ₁ ) = 1, the true number becomes 1, and the average information amount becomes 0.

On the other hand, since the label series data after the division point X ₁ (second group: divided data) includes three labels Z ₁ , Z ₂ , Z ₃ , K = 3, | S ₂ | = 9, freq (Z ₁ , S ₂ ) = 5, freq (Z ₂ , S ₂ ) = 1, and freq (Z ₃ , S ₂ ) = 3.

Therefore, in the example shown in FIG. 9, the average information amount I _X1 of the label series data X ₁ set after the first label as the division point X ₁ is calculated as in the following formula 6.

  Then, the division determination unit 66 evaluates the goodness of the division points by comparing the average information amount of the label series data Z divided by the evaluation unit 666 before setting the division points and at each division point. The evaluation unit 666 evaluates that the best division is to divide the label sequence data Z at a division point where the average information amount is a positive minimum value.

  The division determination unit 66 may be configured to calculate other values such as a chi-square value and a Gini coefficient instead of the average information amount, and the evaluation unit 666 evaluates the calculation result.

  Furthermore, the division determination unit 66 reads the label series data Z of a plurality of actors, and makes the similarity of the label series data of a predetermined period (a preset variable time) before and after the division point between the plurality of actors similar. The degree is calculated by the degree calculator 664. The similarity calculation unit 664 calculates the similarity by, for example, calculating the amount of information on the Cullback library.

In the example shown in FIG. 10A, the average information amount I _X2 of the label series data X ₂ set to the division point X ₂ (division X ₂ ) after the _second label is 0.764. Yes. For example, the similarity calculation unit 664 calculates the similarity of the probability distribution of the labels in a predetermined period after the second labels of the nurse A and the nurse B by the following expression 7.

Here, N represents a predetermined period (a preset variable time). x _i represents the probability distribution φ ^j of the basic action in the upper action z _i = j at time _i .

  As a result of calculating the similarity with the formula 7, the similarity of the probability distribution of the label in the predetermined period after the second label of the nurse A and the nurse B is high, and the upper behavior of the nurse B is clarified. If it is, the evaluation unit 666 evaluates the division point at the same time as the division point of the nurse B as the best division point for the label series data Z of the nurse A.

If nurse A and nurse B have a high degree of similarity in the probability distribution of the label series data, it is possible that nurse A and nurse B are jointly (cooperatively) performing the same superior action. It is done. Specifically, division determination unit 66, when dividing point nurse B that is before the label Z ₃ consecutive (split point X ₈₎ is clear, the similarity of the probability distribution of the label series data to high nurse a, as shown in FIG. 10 (b), to change the split point to split point X _8. That is, the division determination unit 66 determines the division point of the label series data Z of the nurse A having a high similarity to the label series data Z of the nurse B when the higher-order behavior of the nurse B is clear. Is adjusted to the same time as the division point of nurse B.

  Note that the similarity calculation unit 664 may be configured to calculate the similarity based on, for example, the Euclidean distance instead of the information amount of the cullback library.

  The evaluation unit 666 ultimately calculates an average amount of information calculated by dividing the label series data Z of the actor, and the actor and the Both similarities of the label series data Z of other actors are used.

That is, the evaluation unit 666 sets the sum of the average information amount and the Cullback library information amount as the evaluation value E _xn of the division point Xn, as illustrated in the following Expression 8.

Here, α is a coefficient set in advance and satisfies the condition of 0 ≦ α ≦ 1. Then, the evaluation unit 666 evaluates the division point Xn at which the evaluation value E _xn calculated by Equation 8 is the minimum positive value as the best division point.

  In addition, the division determination unit 66 changes the time of the division points in chronological order, calculates the average information amount and the similarity at each division point, respectively, and the sum of the average information amount and the similarity as illustrated in Eq. Each evaluation value is calculated. The division determination unit 66 stores information including the time corresponding to each division point for which the evaluation value has been calculated in the division time DB 57.

FIG. 11 is a chart showing a processing example for sequentially changing the division points and specifying the first best division point. In the processing example shown in FIG. 11, α = 0 (E _xn = I _xn ) is set. In addition, the business flow information includes the upper action names, the upper action order, and the upper action number (= 3) of “send delivery”, “decubitus round”, and “operate sending”.

First, as shown in FIG. 11 (a), the division determination unit 66 assigns “offer” to the first label Z ₂ according to the business flow information for the label series data Z that has not been divided, and Z ₂ The average information amount I _x0 is calculated by the average information amount calculation unit 662 when “decubitus round” is assigned to the label Z ₁ subsequent to “ ₁ ” and “operating feed” is assigned to the label Z ₃ subsequent to Z ₁ . The average information amount I _x0 is 1.485.

Next, as illustrated in FIG. 11B, the division determination unit 66 sets the point after the first label Z ₂ as a division point X ₁ (division X ₁ ), and calculates an average information amount for the divided label series data Z. I _X1 is calculated by the average information amount calculation unit 662. The average information amount I _X1 is 1.216.

Next, as illustrated in FIG. 11C, the division determination unit 66 sets the second label Z ₂ after the division point X ₂ (division X ₂ ) and performs average information on the divided label series data Z. The amount I _X2 is calculated by the average information amount calculation unit 662. The average information amount I _X2 is 0.764.

Then, the division determination unit 66 changes the division points in time series order, and the division label X ₉ (division X ₉ ) follows the ninth label Z ₃ shown in FIG. On the other hand, the average information amount I _X9 is calculated by the average information amount calculation unit 662. The average information amount I _X9 is 1.292.

The division determination unit 66 changes the division point in time series order and calculates the average information amount I _Xn at all the division points, and then divides the label series data Z at the division point at which the average information amount I _Xn becomes a positive minimum value. The evaluation unit 666 evaluates that the division is the best.

In the average information amount I _X0 to the average information amount I _X9 , the average information amount I _X2 is a positive minimum value. That is, in the processing example shown in FIG. 11, the division point X ₂ (division X ₂ ) shown in FIG. 11C is the first best division point.

Division determination unit 66 also performs the respective divided data processing similar to that shown in FIG. 11, while the evaluation value E _xn is a positive value continues to change the split point, the evaluation value E _xn is Continue to select the smallest split point. That is, the division determination unit 66 continues to change the division point until the end condition is satisfied such that the divided data is composed of only the same label or the evaluation value _Exn is not improved (substantially does not change), The evaluation value _{Ex n} and the time corresponding to each division point are continuously stored in the division time DB 57.

  Then, the division determination unit 66 ends the division of the label series data Z when the number of divisions of the label series data Z (the number of divided data) matches the number of higher actions. For example, when the number of upper actions is 3, the division determination unit 66 determines the two best division points until the number of divided data obtained by dividing the label series data Z becomes 3 (label series data Z is 3). Repeat splitting until it is split into two.

  FIG. 12 is a chart showing an example of processing for recursively changing the division points and specifying the second best division point. The process example shown in FIG. 12A specifies the first best division point by the process example shown in FIG.

As shown in FIG. 12 (b), label series data Z, when divided by the dividing point X _2, is divided into the divided data S1 and the divided data S2. As described above, the division determination unit 66 continues to change the division point while the evaluation value E _xn is a positive value, and continues to select the division point at which the evaluation value E _xn is minimum. Here, the divided data S1 is because it is composed of only the same label Z _2, there is no position (time) to be newly set the split point. Therefore, the division determination unit 66 does not divide the divided data S1 any more.

On the other hand, the divided data S2 includes different types of labels. Further, the number of divided data obtained by dividing the label series data Z is smaller than the number of upper actions. Therefore, the division determination unit 66 performs a process of dividing the divided data S2 again. The evaluation value _{Exn at} each division point that can be set for the division data S2 is calculated as shown in the following equations 9-15.

The split point _{X 8} is second best split point of the evaluation value _E x26 = 0.39 which is the minimum value is calculated.

  The determination unit 668 sets the time of each of the division points evaluated by the evaluation unit 666 as the best division point as the final division time for the label series data Z. Then, the determination unit 668 determines the upper action (divided data) of the upper action number obtained by dividing the label series data Z at the final division time. Also, the determination unit 668 outputs the final division time (division time data) to the name assigning unit 67.

  As a specific example, as shown in FIG. 12C, the determination unit 668 uses the times t1 and t2 of the division points evaluated by the evaluation unit 666 as the best division points, as the final division times for the label sequence data Z. And Then, the determination unit 668 determines the upper action (divided data) of the upper action number obtained by dividing the label series data Z into three by the final division times t1 and t2. Also, the determination unit 668 outputs the final division times t1 and t2 to the name assigning unit 67.

  Next, the process which the process part 6 gives and manages the name of a high-order action to each division | segmentation data is demonstrated. First, the name assigning unit 67 reads the business flow information from the business flow DB 58 and receives the final division times t1 and t2 from the determination unit 668. Then, the name assigning unit 67 assigns a higher action name to each of the higher actions (or periods corresponding to the divided data) in accordance with the higher action order according to the division times t1 and t2 and the business flow information. Information (such as label series data Z assigned a higher action name) indicating the higher action assigned a name is output to the fourth management unit 68.

  Specifically, the name assigning unit 67 assigns an upper action name of “send-off” to, for example, the upper action up to the division time t1, and “the pressure ulcer roundabout” for the upper action from the division time t1 to the division time t2. ”And an upper action name“ operating ”is assigned to the upper actions after the division time t2.

  The fourth management unit 68 receives information indicating a higher level action given a higher level action name, and stores the information in the business situation DB 59 for management.

  The output unit 42 receives information indicating the higher level action (business) to which the higher level action name (business name) is assigned via the fourth management unit 68, and displays the higher level action corresponding to the time for each actor.

  FIG. 13 is a display example in which the output unit 42 displays a higher level action given a higher level action name. As illustrated in FIG. 13, the behavior estimation device 1 may be configured such that the output unit 42 displays the result of comparing the scheduled schedule with the actual business situation for each actor.

  Next, processing in which the processing unit 6 detects a sudden event from the label series data Z will be described. The sudden event detection unit 69 accepts the label series data Z stored in the provisional higher order action DB 56 and the business flow information stored in the business flow DB 58. Further, the sudden event detection unit 69 determines whether or not the labels included in the label series data Z are arranged in an order different from the higher order action order included in the business flow information. Then, when it is determined that the labels included in the label series data Z are arranged in an order different from the higher-order action order, the unexpected event detection unit 69 detects the labels arranged in the different order as an unexpected event. .

For example, as shown in FIG. 8, a label division determining unit 66 assigns the "transference" in Z ₂ which is the first label, assign the "bed sores rounds" to Z ₁ is a label followed by Z _2, followed by Z ₁ When “operating” is assigned to Z ₃ , the sudden event detection unit 69 detects Z ₃ arranged between consecutive Z ₁ as a sudden event.

  When the sudden event detection unit 69 detects a sudden event, the sudden event detection unit 69 outputs information indicating the sudden event to the output unit 42. When the output unit 42 receives information indicating the sudden event, the output unit 42 displays information indicating the occurrence of the sudden event, for example.

  In addition, the sudden event detection unit 69 detects whether or not a label different from the label corresponding to the higher level action is included during the operation period of the higher level action given by the name giving unit 67. May be configured.

  Next, the overall operation of the behavior estimation apparatus 1 will be described. First, the operation of the mobile terminal 2 will be described with reference to FIG. 14, and then the operation of the management server 4 will be described with reference to FIGS.

  FIG. 14 is a flowchart showing the operation of the mobile terminal 2. As shown in FIG. 14, the portable terminal 2 detects an acceleration (see FIG. 3A) that changes according to the action of the actor by the detection unit 24 in step 100 (S <b> 100), and acts by the communication unit 26. The position of the person (access point information: see FIG. 3B) is detected.

  In step 102 (S102), the portable terminal 2 transmits the information detected in the process of S100 to the management server 4 as detection information (acceleration data and position data).

  In addition, the portable terminal 2 repeats the processing of S100 and S102 every fixed time (for example, 10 mS).

  FIG. 15 is a flowchart illustrating an operation performed by the management server 4 when the basic operation is estimated. As illustrated in FIG. 15, in step 200 (S200), the management server 4 receives the detection information (acceleration data and position data) transmitted from the mobile terminal 2 by the communication unit 40.

  In step 202 (S202), the management server 4 stores the detection information. Specifically, the first management unit 61 stores acceleration data in the acceleration DB 51 and stores position data in the position DB 52.

  In step 204 (S204), the first estimation unit 62 reads the detection information. Specifically, the first estimation unit 62 reads acceleration data and position data via the first management unit 61.

  In step 206 (S206), the first estimation unit 62 reads the estimation rule 53 (FIG. 4).

  In step 208 (S208), the first estimation unit 62 estimates basic motion (and basic motion sequence data w) from the acceleration DB 51 and the position DB 52 (FIG. 5).

  In step 210 (S210), the second management unit 63 stores the basic motion sequence data w as basic motion data in the basic motion DB.

  FIG. 16 is a flowchart showing an operation performed by the management server 4 using the basic operation data in order to display the business situation. As shown in FIG. 16, in step 300 (S300), the management server 4 reads basic motion sequence data w from the basic motion DB 54 as basic motion data.

  In step 302 (S302), the second estimation unit 64 estimates the provisional higher level action (label) using the basic motion DB 54 and the frequency distribution DB 55.

  In step 304 (S304), the third management unit 65 stores the label series data Z as label data in the provisional higher-level action DB 56.

  In step 306 (S306), the division determination unit 66 reads the business flow information from the business flow DB 58.

  In step 308 (S308), the division determination unit 66 sets a division point in the label series data Z using the business flow information and the provisional higher-level action DB 56.

  In step 310 (S310), the division determination unit 66 calculates the average information amount for the label series data Z by the average information amount calculation unit 662.

  In step 312 (S312), the division determination unit 66 compares the average information amount of the label series data Z (evaluates the goodness of the division points) by the evaluation unit 666.

  In step 314 (S314), the division determination unit 66 uses the similarity calculation unit 664 to calculate the similarity of the label series data for a predetermined period before and after the division point among a plurality of actors.

  In step 316 (S316), the division determination unit 66 evaluates the goodness of the division points using both the average information amount and the similarity.

  In step 318 (S318), the division determination unit 66 determines whether or not the evaluation has been completed for all candidate points (dividable points) that are candidates for division points. If the division determination unit 66 determines that the evaluation has been completed for all candidate points (S318: Yes), the process proceeds to S320. If the division determination unit 66 determines that the evaluation has not been completed for all candidate points (S318: No), the process proceeds to S308.

  In step 320 (S320), the division determination unit 66 determines the best division point.

  In step 322 (S322), the division determination unit 66 determines the necessity of re-division of the divided data depending on whether or not the end condition for ending the division is satisfied. If the division determination unit 66 determines that the end condition is not satisfied and that re-division is necessary (S322: Yes), the division determination unit 66 proceeds to the process of S308. If the division determination unit 66 determines that the termination condition is satisfied and there is no need for re-division (S322: No), the division determination unit 66 proceeds to the process of S324.

  In step 324 (S324), the division determination unit 66 stores the data corresponding to the final division point in the division time DB 57 as the division time data. Specifically, the division determination unit 66 stores the division times t1, t2, and the like illustrated in FIG.

  In step 326 (S326), the name assigning unit 67 assigns a higher action name (business name) to each higher action in accordance with the higher action order according to the division times t1 and t2 and the business flow information.

  In step 328 (S328), the fourth management unit 68 stores information indicating the higher level action given the higher level action name in the task status DB 59 as the task status data.

  In step 330 (S330), the output unit 42 receives information indicating the higher level action (business) given the higher level action name (business name) via the fourth management unit 68, and sets the time for the higher level action for each actor. And display it as a business situation.

  Note that the behavior estimation apparatus 1 of the present embodiment includes each unit included in the processing unit 6 (first management unit 61, first estimation unit 62, second management unit 63, second estimation unit 64, third management unit 65, division The determination unit 66, the name assignment unit 67, the fourth management unit 68, and the sudden event detection unit 69) may be configured as program modules.

  Moreover, although the action estimation apparatus 1 of this embodiment demonstrated as an example the case where it comprised as a client server system which has the some portable terminal 2 and the management server 4, it is not limited to this structure. For example, the behavior estimation apparatus 1 according to the present embodiment may have a configuration in which the function of the management server 4 is distributed to a plurality of mobile terminals 2 and does not include a PC that becomes the management server 4. Moreover, the structure which the mobile terminal 2 has all the functions of the management server 4 may be sufficient as the action estimation apparatus 1 of this embodiment.

  The communication unit 40 of the management server 4 also functions as a change information acquisition unit that acquires change information that changes according to the behavior of the actor. In other words, the management server 4 may be a behavior estimation device.

  According to the embodiment described above, it is possible to estimate and assign a complex higher level action consisting of a combination of basic actions to the estimated basic actions.

  Moreover, although embodiment of this invention was demonstrated by the several combination, these embodiment was shown as an example and is not intending limiting the range of invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

DESCRIPTION OF SYMBOLS 1 Action estimation apparatus 2 Portable terminal 20 Control part 22 Operation part 24 Detection part 26 Communication part 28 Storage part 3 Network 30 Access point 4 Management server 40 Communication part 42 Output part 5 Storage part 51 Acceleration DB
52 Position DB
53 Estimation rules 54 Basic operation DB
55 Frequency distribution DB
56 Provisional upper action DB
57 Division time DB
58 Business Flow DB
59 Business Status DB
6 processing unit 61 first management unit 62 first estimation unit 63 second management unit 64 second estimation unit 65 third management unit 66 division determination unit 660 division unit 662 average information amount calculation unit 664 similarity calculation unit 666 evaluation unit 668 Determination unit 67 Name assignment unit 68 Fourth management unit 69 Sudden event detection unit

Claims (7)

A change information acquisition unit that acquires change information that changes according to the behavior of the actor;
A first estimation unit that estimates a plurality of basic motions arranged in a time series of the actor based on the change information;
A storage unit that stores a business flow including k (k is an integer of 2 or more) in a predetermined order in a higher order action constituted by a combination of basic actions;
Based on the probability distribution of the higher-level actions of a plurality of basic actions included in a predetermined period, the higher-order actions configured by combining the basic actions estimated by the first estimation unit are provisionally higher for each basic action. A second estimator for estimating the behavior;
Based on the business flow, the plurality of temporary higher-order actions arranged in time series are divided into k groups arranged in time series, and each of the groups includes the higher-order actions included in the business flow according to the order. A division determination unit for assigning
A behavior estimation apparatus. The business flow includes k upper action names associated with the order,
The behavior estimation apparatus according to claim 1, further comprising: a name assigning unit that assigns the upper action name in accordance with the order to each of the upper actions assigned by the division determination unit. The division determination unit
A division unit that sequentially divides the plurality of provisional upper actions arranged in time series into a plurality of provisional upper action columns including one or more provisional upper actions;
An average information amount calculation unit that calculates an average information amount for each column of the provisional higher level action divided by the division unit;
An evaluation unit that evaluates that the division in which the average information amount calculated by the average information amount calculation unit is a positive minimum value is the best division;
A determination unit that determines the k higher-order actions according to a result of the evaluation performed by the evaluation unit as being the best division;
The behavior estimation apparatus according to claim 1, comprising: The division determination unit
A similarity calculation unit that calculates the similarity of the column of the provisional higher-order actions estimated by the second estimation unit for each basic action for each of a plurality of actors;
A determination unit that determines the k top actions by comparing the similarities calculated by the similarity calculation unit;
The behavior estimation apparatus according to claim 1, comprising: The division determination unit
A division unit that sequentially divides the plurality of provisional upper actions arranged in time series into a plurality of provisional upper action columns including one or more provisional upper actions;
An average information amount calculation unit that calculates an average information amount for each column of the provisional higher level action divided by the division unit;
A similarity calculation unit that calculates the similarity of the column of the provisional higher-order actions estimated by the second estimation unit for each basic action for each of a plurality of actors;
An evaluation unit that evaluates that the division in which the sum of the average information amount and the similarity is a positive minimum value is the best division;
A determination unit that determines the k higher-order actions according to a result of the evaluation performed by the evaluation unit as being the best division;
The behavior estimation apparatus according to claim 1, comprising:   A sudden event detection unit that detects the provisional high-order behavior estimated in an order different from the order as a sudden event when the second estimation unit estimates the provisional high-level behavior in an order different from the order; The behavior estimation device according to claim 1.   The behavior estimation apparatus according to claim 2, further comprising: an output unit that outputs, in time series, the superior behavior determined by the division determination unit using the superior behavior name assigned by the name assigning unit.

Images