JP4848900B2 - POSITION INFORMATION ESTIMATION DEVICE, POSITION INFORMATION ESTIMATION METHOD, AND POSITION INFORMATION ESTIMATION PROGRAM - Google Patents

Description

  The present invention relates to a position information estimation method, a position information estimation apparatus, and a position information estimation program for estimating the position of a user using a sensor device, and in particular, even if a user has left the sensor device in place. The present invention relates to a usable location information estimation method, a location information estimation device, and a location information estimation program.

  In recent years, a service has been proposed in which the position of a user is detected using a sensor device and the detected position is used.

  Patent Document 1 describes a communication service method in which position information indicating a user's current position is treated as presence information. In the communication service method described in Patent Document 1, position information as presence information is acquired from a position detection sensor such as a wireless LAN, RFID, or GPS or PHS mounted on a portable terminal carried by a user.

  Patent Document 2 describes a system that detects that a user does not carry a sensor device. The system described in Patent Literature 2 determines that the user is in an abnormal state in which the user does not carry a sensor device when the user's position information detected by a plurality of sensors contradicts.

Non-Patent Document 1 describes a method of using SIP as a presence protocol, which is a protocol used for presence services. Non-Patent Document 2 describes a sensor information acquisition method using SIP. Non-Patent Document 3 describes a hidden Markov model (HMM), which is an example of a probability model.
JP 2004-328309 A (paragraphs 0026-0031) Japanese Patent Laying-Open No. 2005-38269 (paragraphs 0019-0080) J. Rosenberg, "A Presence Event Package for the Session Initiation Protocol (SIP)", RFC 3856, 2004 Yasuaki Chimura and Toshifumi Murata, “Revised SIP Textbook”, Impress, 2004, p. 205-218 LR Rabiner, and BH Juang, "An Introduction to Hidden Markov Models", IEEE ASSP Magazine, 1986, Vol. 3, No. 1, pp. 4-16

  The position of the user is indicated by using a user terminal (for example, a portable terminal that transmits position information or an ID transmitter such as an RFID tag) equipped with a position detection sensor or the like as described in Patent Document 1. The method for obtaining position information is insufficient in the following points.

  This method is based on the premise that the user carries a user terminal that transmits position information. For this reason, when the user does not wear the user terminal, position information that does not accurately indicate the position of the user is provided.

  Further, the method using a plurality of sensors (for example, a mobile phone, an IC card, and a human sensor) as described in Patent Document 2 is insufficient in the following points.

  In this method, a single sensor cannot detect that the user is not carrying the sensor device. Furthermore, when the user does not carry the sensor device and any sensor information is incorrect, there is no standard for determining which sensor information is reliable.

  Further, in the methods described in Patent Document 1 and Patent Document 2, it is impossible to examine where the user is when the user does not carry the sensor device.

  An object of the present invention is to provide a position information estimation apparatus, a position information estimation method, and a position information estimation program capable of estimating the position of a user even when the user does not carry a sensor device. It is to be.

  In order to achieve the above object, a position information estimation apparatus according to the present invention is a position information estimation apparatus that estimates the position of a user based on sensor information correlated with the position of a sensor device carried by the user. The sensor information acquisition means for acquiring the sensor information, the probability model that the user and the sensor device can be at different positions, the position of the user is represented, the probability model, Position estimation means for estimating the position of the user based on the sensor information acquired by the sensor information acquisition means.

  The position information estimation method of the present invention is a position information estimation method performed by a position information estimation apparatus that estimates the position of a user based on sensor information correlated with the position of a sensor device carried by the user. A sensor information acquisition step for acquiring the sensor information, and a probability model in which the user and the sensor device can be at different positions, and the position of the user is expressed, and the probability model and the sensor And a position estimating step for estimating the position of the user based on the information.

  The position information estimation program of the present invention is a position that causes a computer to execute position information estimation processing for estimating the position of the user based on sensor information correlated with the position of a sensor device carried by the user. In the information estimation program, the sensor information acquisition process for acquiring the sensor information, and the probability model that the user and the sensor device can be at different positions, the position of the user is expressed, and the probability A position information estimation process including a position estimation process for estimating the position of the user based on the model and the sensor information is executed by a computer.

  According to the above invention, the position of the user is estimated based on the probability model that the user and the sensor device can be at different positions and the sensor information. For this reason, the user's position can be estimated taking into account the possibility that the user and the sensor device are not in the same location, such as when the user misplaces the sensor device, and the user's position is estimated with high accuracy. It becomes possible to do.

  The position estimation means includes a sensor output parameter indicating a correlation between the sensor information and the position of the sensor device, and a misplacement parameter indicating a correlation between the position of the sensor device and the position of the user. Parameter storage means for storing as a probability model parameter corresponding to the sensor, and the sensor information acquisition means acquires the position of the sensor device and the position of the user as non-observable variables that are information that cannot be observed from the outside. It is desirable to have non-observable variable deriving means derived based on the information and the probability model parameters stored in the parameter storage means.

  The sensor information acquisition unit further acquires an observable variable having a correlation with the position of the user, and the parameter storage unit further indicates a correlation between the observable variable and the position of the user. An observation variable output parameter is stored as the probability model parameter, and the non-observable variable derivation unit includes the sensor information and observable variable acquired by the sensor information acquisition unit, and the probability model parameter stored in the parameter storage unit. It is desirable to derive the position of the sensor device and the position of the user based on the above.

  The position estimation means further includes history storage means for storing the estimated position of the sensor device and the estimated position of the user according to a time series, and the non-observable variable derivation means includes at least the history storage. Based on the position of the sensor device and the position of the user stored in the means, the sensor information acquired by the sensor information acquisition means, and the probability model parameter stored in the parameter storage means, the current sensor device And the current position of the user is preferably derived.

  The probabilistic model is a hidden Markov model or a dynamic Bayesian network model that estimates the non-observable variable that changes with time based on at least the sensor information, and the parameter storage means includes the position of the sensor device at each time and Parameters for formulating the user's position as each internal state of the hidden Markov model or the dynamic Bayesian network model are stored as the probability model parameters, and the non-observable variable deriving means includes at least the It is desirable to derive the current position of the sensor device and the current position of the user based on the sensor information acquired by the sensor information acquisition means and the parameters stored in the parameter storage means.

Sensor wherein the parameter storage means, as said parameter, indicating at least the probability of the sensor information x _t is output if the position of the sensor device is shifted to the internal state s _t corresponding to the time t at time t An output probability and a misplacement probability that is a probability that the user misplaces the sensor device are stored, and the non-observable variable deriving unit stores the sensor information obtained by the sensor information obtaining unit and the parameter storage unit. Based on the stored sensor output probability and misplacement probability, the current position of the current sensor device and the current position are obtained by determining the probability that each of the sensor device and the user is in each internal state at the current time. It is desirable to derive the position of the user.

  The misplacement probability is preferably determined in relation to time or the internal state, or the sensor information.

  The sensor information acquisition means acquires an observable variable having a correlation with the position of the user, and the misplacement probability is determined in relation to time or the internal state, the sensor information, or the observable variable. It is desirable.

The sensor information acquisition unit acquires an observable variable y having a correlation with the position of the user, and the parameter storage unit further sets the sensor device and the user as an internal state s _i first as the parameter. An initial existence probability indicating the probability of being in a state (i is a natural number, and when each of the internal states is expressed as internal states s ₁ to s _j , 1 <i ≦ j); state s _t-1 from the state transition probability is the probability of transition to the internal state s _t (t is a natural number corresponding to the time of observation: 1 <t) and the in the sensor device is shifted to the internal state s _t stores, and observable variable output probability indicating the probability that the observable variable y _t is output, the non-observable variables deriving means, said parameter storage means storing initial existence probability, the state transition probability, observability Variable output probability, sensor output Based on the force probability and the misplacement probability, the position of the current sensor device and the position of the current user are obtained by determining the probability that each of the sensor device and the user is in each internal state at the current time. It is desirable to derive

The non-observable variables deriving means, as the state transition probability, the internal state at the previous time s _t-1 and the conditional probability P for observability variable y _t at the current time _{(s t | s t-1} , y t) It is desirable to use

The non-observable variables deriving means, as said probability model, the hidden Markov model, upon the output symbols of the sensor information, sequence {x ₁ of the sensor information x _t at the current time t, x _2, .., X _t } are observed, the posterior probabilities indicating the probability that each of the sensor device and the user is in each internal state s _i are _expressed as conditional probabilities P (s _i | x ₁ , x ₂ , .., X _t ) are preferably obtained using a forward algorithm for the hidden Markov model.

The non-observable variable deriving means has a random variable representing the position of the user at time t as π _t and a random variable representing the position of the sensor device as λ _t, and the position of the user and the sensor It is desirable to perform calculation by regarding a set of device positions (π _t , λ _t ) as one variable.

The parameter storage means stores a misplacement probability p _drop which is a probability that the user will misplace the sensor device, and the non-observable variable derivation means obtains a random variable representing the position of the user at time t as π _t, wherein the random variable representing the sensor device positions the lambda _t, the sensor position conditional probability regarding P devices _{(λ t | λ t-1} , π t-1, π t) be calculated by the following equation Is desirable.

Preferably, the p _drop is not a constant but depends on the π _t-1 or the π _t or other sensor information.

The sensor information acquisition means acquires an observable variable having a correlation with the position of the user, and the p _drop is not a constant, but the π _t-1 or the π _t or other sensor information or the observable It is desirable to depend on variables.

The probability model is a hidden Markov model that estimates the non-observable variable that changes with time based on at least the sensor information, and the parameter storage means corresponds to the position of the sensor device at time t. The sensor output probability indicating the probability that the sensor information is output when transitioning to the internal state of the hidden Markov model, and the probability that the sensor device and the user are initially in the internal state s _i (i is a natural number) There is an initial existence probability indicating that each internal state satisfies 1 <i ≦ j when the internal states are represented as internal states s ₁ to s _j , and the user determines from the internal state s _t-1 to the internal state s a state transition probability (t is a natural number corresponding to the observed time: 1 <t), which is a probability of transition to _t , and the non-observable variable deriving means stores each state s at time 0 _From the initial existence probability I _{i of i} , the state transition probability, the sensor output probability, the sensor information at the current time, and the calculation result of the forward probability at the previous time, the user at the time t has the internal state s _i , the sensor Find the forward probability f _{i, k} (t) that the device is in s _k by the following equation:

Further, using the determined forward probability f _{i, k} (t), the position of the user is in the state s _i when the sensor information sequence is observed at time t, and the sensor device is in the state s _k . The posterior probability P _{i, k} (t) indicating a certain probability, the posterior probability P _i (t) indicating the probability that the user's position is in the state s _i, and the probability that the sensor device is in each state s _k It is desirable to obtain the posterior probability P _k (t) to be shown using the following equation.

The sensor information acquisition means acquires an observable variable y correlated with the position of the user, and the probability model sets the non-observable variable that changes with time to at least the sensor information and the observable variable y. a hidden Markov model that estimates based, said parameter storage means, a sensor that indicates the probability that the sensor information is output if the position of the sensor device is shifted to the internal state s _t of the hidden Markov model at time t and output probability, and observability variable output probability indicating the probability of the observable variable is output when said user location at time t is shifted to the internal state s _t, first the sensor device and the user Is the internal state s _i (i is a natural number, and when each internal state is expressed as the internal state s ₁ to s _j , 1 <i ≦ j is satisfied) Storing an initial existence probability and a state transition probability (t is a natural number corresponding to the observed time: 1 <t), which is a probability that the user transitions from the internal state s _t-1 to the internal state s _t ; The non-observable variable deriving means includes initial existence probability I _i of each state s _i at time 0, state transition probability, sensor output probability, observable variable output probability, sensor information x and observable variable y at the current time, And the forward probability f _{i, k} (t) where the user is at s _i and the sensor device is at s _k at time t from the calculation result of the forward probability at the previous time, and

Further, using the determined forward probability f _{i, k} (t), the position of the user is in the state s _i when the sequence of sensor information is observed at time t, and the sensor device is in the state s _k . A posteriori probability P _{i, k} (t) indicating the probability, a posteriori probability P _i (t) indicating the probability that the user's position is in the state s _i , and the sensor device is in each state s _k It is desirable to obtain the posterior probability P _k (t) indicating the probability using the following equation.

  The position estimation means derives an incidental probability representing a ratio of the user carrying the sensor device using the probability of the position of the user and the position of the sensor device estimated based on the probability model. It is desirable to further have an incidental probability deriving means.

The incidental probability deriving means has π _t = λ _t when the user position probability π _t and the sensor device position probability λ _{t at} time _t are obtained by the non-observable variable derivation means. It is desirable to obtain the posterior probability P _r (π _t = λ _t | x ₁ , x ₂ ,..., X _t ) of the event.

  According to the present invention, it is possible to estimate a user position more accurately in consideration of misplacement of a sensor device. The reason is that the user's position is estimated based on a probabilistic model that also considers the possibility that the user and the sensor device are not in the same place, such as when the user misplaces the sensor device.

  Next, embodiments of the present invention will be described in detail with reference to the drawings.

(First embodiment)
FIG. 1 is a block diagram showing a position information estimation system including a position information estimation apparatus according to the first embodiment of the present invention.

  In FIG. 1, the position information estimation system is also referred to as a user terminal 11 that is carried by a user 1 and is a sensing target and a position of the user terminal 11 (hereinafter also referred to as “user terminal position” or “sensor device position”). .) And the position information estimation server 2 for estimating the position of the user 1 (hereinafter referred to as “user position”) using the position of the user terminal 11 detected by the sensor system 211. And a service providing server 3 that provides services such as information provision using the user position estimated by the position information estimating server 2.

  The position information estimation server 2 includes a position estimation unit 20 and a sensor information acquisition unit 21. The position estimation unit 20 includes a parameter storage unit 22, a non-observable variable estimation unit 23, and an output signal storage unit 24. Although only one user terminal 11 is shown in FIG. 1, a plurality of user terminals 11 may exist.

  The user terminal 11 is an example of a sensor device. For example, an ID transmitter such as an RFID tag, a portable terminal that establishes a wireless link with a specific base station, or a portable terminal that transmits position information by itself using a GPS or the like. It is a terminal device, and is a portable terminal device that can cause the sensor system 211 to detect sensor information indicating the position of the user terminal 11.

  The sensor system 211 is a system that detects the position of the user terminal 11 and outputs sensor information indicating the detected position to the position information estimation server 2. Specifically, the sensor system 211 performs information processing such as a personal computer that operates according to a program. Device.

  The sensor system 211 may be a presence service system that provides information according to a presence event package of a session initiation protocol (SIP).

  The presence service system is a system that performs presence service that accepts presence information and distributes the presence information to watchers.

  A method of using SIP for the presence protocol, which is a protocol used for the presence service, is described in Non-Patent Document 1, for example. As a sensor information acquisition method using SIP, a method widely known in Non-Patent Document 2 or the like can be used. Non-Patent Document 2 describes a presence service mechanism (configuration), usage of a subscription / notification method (sequence example) for realizing a presence service, and a procedure thereof as a method for realizing a presence server using SIP. Has been.

  If the sensor system 211 is realized using such a presence service, for example, an ID transmitter such as an RFID tag, a portable terminal that establishes a link wirelessly with a specific base station, or a GPS is used to transmit position information by itself. By providing the user with a user terminal 11 such as a portable terminal, information (sensor information) indicating a position associated with the user terminal 11 can be detected.

  Here, the sensor information detected by the sensor system 211 (positions that can be taken by the user terminal 11) can be assumed to be {meeting room 1, meeting room 2, cafeteria, out of service area}, etc., taking the office environment as an example. Out of service area means the output of the sensor system 211 when the sensor system 211 cannot detect the sensor information from the user terminal 11. The sensor system 211 may collect sensor information periodically and store the sensor information in time series. Further, the sensor system 211 may be provided inside the position information estimation server 2.

  The service providing server 3 is a server device that provides information or notifies the user position according to the user position estimated by the position information estimation server 2, specifically a personal computer that operates according to a program, etc. And is realized by a CPU that operates according to a program. Further, the service providing server 3 may be provided inside the position information estimation server 2.

  Next, the configuration of the position information estimation server 2 will be described.

  The position information estimation server 2 is an example of a position information estimation apparatus, and is a server apparatus that estimates a user's position, and is specifically an information processing apparatus such as a personal computer that operates according to a program.

  The sensor information acquisition unit 21 collects sensor information and passes the sensor information to the position estimation unit 20 (specifically, the non-observable variable estimation unit 23). For example, the sensor information acquisition unit 21 makes an inquiry to the sensor system 211, periodically collects sensor information indicating the position of the user terminal 11, and stores the collected sensor information in the output symbol storage unit 24 according to the time axis. Also good.

  The position estimation unit 20 expresses (formulates) the position of the user 1 with a probability model that the user 1 and the user terminal 11 can be at different positions, and the probability model and the sensor information acquisition unit 21. The position of the user 1 and the position of the user terminal 11 are estimated on the basis of the acquired sensor information.

  The parameter storage unit 22 is a parameter (probability model parameter (for example, for example) used for deriving the user position and the user terminal position based on sensor information correlated with the user position and the user terminal position to be estimated. The sensor output parameter indicating the correlation between the sensor information and the user terminal 11 and the misplaced parameter indicating the correlation between the position of the user terminal 11 and the position of the user 1) are stored. The parameters may be stored in advance in the parameter storage unit 22, or actual data may be input and parameters obtained by learning from the actual data may be stored in the parameter storage unit 22 as needed.

  The output symbol storage unit 24 stores sensor information that is an output symbol in HMM (Hidden Markov Models) along a time series. Further, the output symbol storage unit 24 may store past estimation results (past user position and user terminal position) used for estimation of the current position in time series, and further for parameter learning. Actual data to be used may be stored.

  The non-observable variable estimation unit 23 is based on the sensor information collected by the sensor information acquisition unit 21, the parameters stored in the parameter storage unit 22, and the previously derived user position and user terminal position. The probability that the user 1 and the user terminal 11 are at each position (a random variable indicating a position that can be taken by the user terminal 11 in association with each internal state of the HMM) when the column of sensor information is observed ( A posteriori probability) is determined using a forward probability for the HMM (probability model).

  In the present embodiment, sensor information acquisition unit 21 and non-observable variable estimation unit 23 are realized by a CPU that operates according to a program. The parameter storage unit 22 and the output symbol storage unit 24 are realized by a storage device. The program is stored in a storage device provided in the position information estimation server 2.

  Here, a method for estimating the position information of the user in consideration of misplacement of the sensor device (user terminal) will be described.

  The position information estimation server 2, specifically, the non-observable variable estimation unit 23 is a probability model in which the user position and the sensor device position can take different states, which are conventionally treated as being the same, The user position is expressed (formulated), and the user position is estimated based on the probability model.

  The non-observable variable estimation unit 23 derives the user position and the sensor device position, which are estimation targets, as random variables. Specifically, non-observable variable guess. The fixing unit 23 determines the user position and the sensor device position to be estimated by each element of a set of contents (positions) that the sensor device can take (for example, in the case of an office, a set {conference room, living room, cafeteria, return home}). Derived by the random variable to represent.

  The non-observable variable estimation unit 23 converts a random variable, which is information that cannot be observed from the outside (hereinafter referred to as a non-observable variable), into an observable variable (a variable that can be observed from the outside) having a correlation with the non-observable variable. ) And a probability model that the user position and the sensor device position can assume different states.

  It should be noted that the observable variable may include sensor information that is information of detecting an actual sensor device.

  Further, for example, when the user position and the sensor device position have the property of changing over time, the non-observable variable estimation unit 23 determines that the user position and the sensor device position at the current time are It is assumed that there is a correlation between the user position and the sensor device position at an earlier time, and using the user position and sensor device position derived in the past, together with the observable variable, the current user position and sensor device position by the probability model May be derived.

  FIG. 2 is an explanatory diagram for explaining a probability model for estimating the user position and the sensor device position in consideration of misplacement of the sensor device.

  In FIG. 2, non-observable variables (user positions and sensor device positions in the figure) to be estimated and observable variables correlated with them (sensor information in the figure) are shown as ellipses as random variables. ing. Further, in FIG. 2, the fact that there is a correlation between random variables is indicated by an arrow, and the correlation is indicated by a conditional probability.

For example, in the example shown in FIG. 2, the non-observable variables to be estimated are the user position π _t and the sensor device position λ _t (t is a natural number corresponding to time: 1 <t).

  Also, sensor information x is shown as an observable variable having a correlation with the sensor device position.

Further, the user position π _{t at} time t is indicated by a conditional probability P (π _t | π _t-1 ) assuming that it has a correlation with the immediately preceding user position π _t-1 .

The first sensor device position λ ₁ has a correlation with the user position π _1, and the correlation is indicated by a conditional probability P (λ ₁ | π ₁ ).

Also, the position lambda _t of the sensor device at time t, and the user position [pi _t, the sensor device position immediately before lambda _t-1, and, as having user position [pi _t-1 both correlation, conditional probability P ( λ _t | λ _{t −1} , π _t−1 , π _t ).

Further, the sensor information x is indicated by a conditional probability P (x _t | λ _t ) assuming that it has a correlation with the position λ of the sensor device.

  It is assumed that the parameter storage unit 22 stores in advance parameters used for deriving non-observable variables. The parameters mentioned here are, for example, sensor output parameters (conditional probability) indicating the correlation between sensor information and sensor device position, and misplaced parameters (conditional probability) indicating the correlation between sensor device position and user position. ) And a random variable correlated with the non-observable variable (for example, the observable variable, the past user position, and the sensor device position) are information indicating the relative relationship with the non-observable variable. .

  Next, a specific estimation method will be described by taking as an example a case where an HMM is used as a probability model.

  HMM is a probabilistic model with a structure similar to that of a finite state machine, and cannot be observed from the outside. It is a Markov process (a stochastic process that assumes that the appearance probability of a certain symbol depends only on the previous one) and its state. This is a model that expresses a series of output symbols by a combination of output symbols. The HMM is described in Non-Patent Document 3 described above.

When using the HMM, the non-observable variable estimation unit 23 sets the non-observable variables (user position and sensor device position) to be estimated as the internal state of the HMM (hereinafter simply referred to as “state”). deal with. That is, the user position and the sensor device position are an aggregate {s ₁ , s ₂ ,..., S _j } with a state number _j (for example, in the case of an office, the aggregate {s ₁ = conference room) , s ₂ = room, s ₃ = cafeteria, s ₄ = go home}) corresponding to any one state s _i of the (i is a natural number corresponding to each state: 1 <i ≦ j).

  FIG. 3 is an explanatory diagram illustrating an example of parameters in the HMM. FIG. 3 shows an example of parameters when there is a correlation as shown in FIG. These parameters are stored in the parameter storage unit 22. The parameters in the HMM include at least an initial existence probability, a state transition probability, a sensor device existence probability, and a sensor output probability.

The initial existence probability is a probability that the user and the sensor device are in each state s _i at time 0, and is represented by a probability I _i . In this embodiment, it is assumed that the user and the sensor device are in the same state (position) at time 0.

The state transition probability is a probability that the user transitions from the state s _{t -1} to the state s _t .

Existence probability of the sensor device, the state of the user [pi _t-1 = s _t-1 is the state of and the sensor device becomes π _t = s _t from lambda _t-1 = s from _t-1 state lambda _t = a probability that a s _t, is given by the probability that the user misplaced the sensor device (misplaced probability = misplaced parameters). The misplacement probability p _drop is stored in the parameter storage unit 22.

Output probability, when the sensor information was output symbols, a probability that the sensor information x _t is output when the sensor device transits to the state s _t.

The position between the moving probability of the state transition probability refers to a probability of moving to the current position [pi _t from the position [pi _t-1 immediately preceding the user time t, the conditional probability _{P (π t | π t-} 1) Indicated by

The existence probability of the sensor device is conditional on the sensor device position λ _{t at} time t being correlated with the user position π _t and the immediately preceding sensor device position λ _t-1 and user position π _t-1. Probability P (λ _t | λ _t−1 , π _t−1 , π _t ) is indicated.

  Since the sensor device does not move spontaneously, the non-observable variable estimation unit 23 may calculate the existence probability of the sensor device using the following equation (Equation 11).

The output probability indicates a probability that the sensor information x _t is output when the sensor device position λ _t is changed, and is indicated by a conditional probability P (x _t | λ _t ).

Based on the above, when the sensor information is viewed as an output symbol and the sequence {x ₁ , x ₂ ,..., X _t } of the sensor information is observed, the non-observable variable estimation unit 23 Based on the information sequence and the parameters stored in the parameter storage unit 22, the probability P _i (t) that the value of the user position π _t is the state (position) s _i can be obtained.

This probability P _i (t) is also called a posterior probability, and is represented by a conditional probability P (π _t = s _i | x ₁ , x ₂ ,..., X _t ). The probability P _i (t) can be derived by recursively calculating P _i (t−1) using a forward algorithm for the HMM.

Specifically, the non-observable variable estimation unit 23 first determines in which state the user and the sensor device are at time 0 based on the initial existence probability. At this time, it is assumed that the user position π ₁ and the sensor device position λ ₁ are the same.

In addition, the non-observable variable estimation unit 23 determines the state of the user at time t from the state transition probability of the HMM, the presence probability and sensor output probability of the sensor device, the sensor information at the current time, and the past posterior probability. The forward probability f _{i, k} (t) where s _i and the sensor device position is in the state s _k is _obtained by the following equation (Equation 12).

  Here, i represents in which state the user is in time t, and k represents in which state the sensor device is in. Further, m and l are symbols given for convenience to distinguish from i and k at time t, and represent the state of the user position and sensor device position at time t-1.

Next, the non-observable variable estimation unit 23 uses the obtained forward probability f _{i, k} (t), the user position is in the state s _i when the sequence of sensor information is observed at time t, and the sensor posterior probability P _i that the device indicates the probability that the state s _k, a _k (t), and the posterior probability P _i (t) indicating the probability that a state s _i is the user position, the sensor device to each state s _k The posterior probability P _k (t) indicating the probability of being present is calculated using the following equation (Equation 13). In a few 13, P _i, k a (t) p _i, denoted as k (t), represents P _i a (t) p _i (t) and, P _k a (t) p _k (t) and Represents.

  By using such a method, the non-observable variable estimation unit 23 can derive the user position and the sensor device position in the form of random variables.

Furthermore, since the posterior probability P _{i, k} (t) = P (π _t = s _i , λ _t = s _k | x ₁ , x ₂ ,..., X _t ) is obtained by the above calculation, π _t = posterior probability of the event that is λ _t P | can be calculated _{(π t = λ t x 1} , x 2, ···, x t) = Σ s P i, k a (t). That is, it is possible to calculate an incidental probability that is a probability that the user and the sensor device are at the same position (the user is carrying the sensor device).

  Although the parameter storage unit 22 has been described as storing parameters in advance, the parameters can also be obtained by learning using actual data. For example, the non-observable variable estimation unit 23 may use actual sensor information and a record of the stay position, obtain parameters that match the parameters, and store the obtained parameters in the parameter storage unit 22.

  Specifically, when an output symbol (sensor information) and a state at that time (true stay position) are obtained as data for learning, the non-observable variable estimation unit 23 performs maximum likelihood inference (maximum likelihood estimation).

  Further, when only output symbols are obtained as data for learning, the non-observable variable estimation unit 23 may obtain parameters using the Baulm-Welch algorithm. The Baum-Welch algorithm is a kind of learning method called an EM algorithm, which is known as a method for acquiring parameters used in the HMM. The parameter is repeatedly improved, and the likelihood of the parameter increases monotonously with each iteration. It is an algorithm that is guaranteed. The Baum-Welch algorithm is described in Non-Patent Document 3, for example.

  Next, the operation of the present embodiment will be described with reference to FIG. FIG. 4 is a flowchart showing an example of position estimation processing by the position information estimation system of the present embodiment.

  First, the sensor information acquisition unit 21 of the position information estimation server 2 inquires the sensor system 211, periodically receives the sensor information, and stores the received sensor information in the output symbol storage unit 24 (step S21).

  Here, for example, when the service providing server 3 transmits a request message for requesting the position of the user 1 to the position information estimation server 2 (step S31), the non-observable variable estimation unit 23 of the position information estimation server 2 is Then, parameters are read from the parameter storage unit 22 (step S24).

Then, the non-observable variable estimation unit 23 recursively calculates the probability (posterior probability) P _i (t) that the user 1 and the user terminal 11 are in each state s _i at time t using a forward algorithm for the HMM. Is calculated while calculating _Pi (t-1) (step S25).

Then, the non-observable variable estimation unit 23 sends the probability that the user 1 is in each state s _i (posterior probability) and, if necessary, the probability that the user terminal 11 is in each state s _k to the service providing server 3. (The posterior probability) is transmitted, and the service providing server 3 transmits the probability that the user 1 is in each state s _i (posterior probability) and, if necessary, the probability that the user terminal 11 is in each state s _k (posterior probability). ) Is received (step S32).

  As described above, according to the present embodiment, it is possible to estimate the position of the user 1 more accurately in consideration of misplacement of the user terminal 11. That is, it is possible to estimate the position of the user 1 with higher accuracy.

  In addition, the position of the user terminal 11 can be estimated by a single sensor.

  Further, when the position of the user 1 is obtained, a hidden Markov model (probability model) that derives the position of the user 1 based on the past state is used, so that the user 1 does not carry the sensor device. However, the position where the user 1 exists can be obtained with probability.

  In the present embodiment, the sensor information acquisition unit is realized by the sensor information acquisition unit 21. The position estimation unit is realized by the position estimation unit 20. The parameter storage unit is realized by the parameter storage unit 22. The non-observable variable deriving unit is realized by the non-observable variable estimating unit 23. The history storage means is realized by the output symbol storage unit 24.

(Second Embodiment)
Next, a position information estimation system will be described in which the position information estimation server further estimates the probability (incident probability) that the user 1 is carrying the user terminal 11.

  FIG. 5 is a block diagram showing a position information estimation system according to the second exemplary embodiment of the present invention. In FIG. 5, the same components as those shown in FIG. In FIG. 5, the position information estimation server 2 a further includes an incidental probability estimation unit 25 that estimates an incidental probability in addition to the components of the position information estimation server 2 of the first embodiment. Hereinafter, the second embodiment will be described with a focus on differences from the first embodiment.

The incidental probability estimation unit 25 obtains the probability (posterior probability) P _{i, k} (t) obtained by the non-observable variable estimation unit 23 that the user 1 is in the state s _i and the user terminal 11 is in the state s _k. ) To derive the probability that the user 1 is carrying the user terminal 11.

Specifically, incidental probability estimator 25, [pi _t = posteriori probability of an event that is _{λ t P (π t = λ} t | x 1, x 2, ···, x t) = Σ s P i _{, k} (t). That is, the incidental probability estimation unit 25 calculates an incidental probability that is a probability that the user 1 and the user terminal 11 are in the same position (the user 1 is carrying the user terminal 11).

  The position estimation unit 20a includes a parameter storage unit 22, an unobservable variable estimation unit 23, an output symbol storage unit 24, and an incidental probability estimation unit 25.

  FIG. 6 is a flowchart illustrating an example of the position estimation process of the position information estimation system according to the second embodiment. In FIG. 6, the same operations as those shown in FIG. 4 are denoted by the same reference numerals.

After the non-observable variable estimation unit 23 estimates the probability (posterior probability) P _{i, k} (t) that the user 1 is in the state s _i and the user terminal 11 is in the state s _k (step S25). ), The incidental probability estimation unit 25 uses the posterior probability by multiplying the probabilities that the user 1 and the user terminal 11 are in the same position in each state for all the states. The attendant probability that the user 1 is carrying the user terminal 11 is derived, and the attendant probability is provided to the non-observable variable estimation unit 23 (step S26).

Then, the non-observable variable estimation unit 23 transmits, to the service providing server 3, the probability that the user 1 and the user terminal 11 are in each state s _i and each state s _k (posterior probability) and incidental probability, and the service The providing server 3 receives the probability (posterior probability) and incidental probability that the user 1 and the user terminal 11 are in each state s _i and each state s _k (step S32).

  As described above, according to the present embodiment, it is possible to obtain the ratio (incident probability) that the user 1 is carrying the user terminal 11. Moreover, according to this Embodiment, it becomes possible to derive an incidental probability from a single sensor.

  In the present embodiment, the incidental probability derivation means is realized by the incidental probability estimation unit 25.

(Third embodiment)
Next, as an observable variable correlated with the user position, in addition to the sensor information in the second embodiment, other observable variables (for example, other sensor information, schedule table, time zone variables, etc.) An example of a position information estimation system that uses the above will be described.

  FIG. 7 is a block diagram showing a position information estimation system according to the third exemplary embodiment of the present invention. In FIG. 7, the same components as those shown in FIG. In the following, the third embodiment will be described focusing on differences from the second embodiment.

  The position information estimation system shown in FIG. 7 includes a schedule table system 211-1 and a time server 211-2 in addition to the sensor system 211, and includes a position information estimation server 2b instead of the position information estimation server 2a. The position information estimation server 2b includes a sensor information acquisition unit 21a, an output signal storage unit 24a, a parameter storage unit 22a, a non-observable variable estimation unit 23a, and an incidental probability estimation unit 25.

  The schedule table system 211-1 is a system that manages schedule information in units of users and outputs a schedule of at least a certain user at the current time to the position information estimation server 2 b. The schedule information is information indicating a schedule related to the position of the user. For example, the date of the schedule, the start time, the end time, a requirement representing the contents of the schedule, and the location of the schedule may be included.

  When the schedule table system 211-1 receives an inquiry about a schedule of a certain user at the current time, the schedule table system 211-1 refers to the schedule information of the corresponding user and at least shows a schedule (hereinafter simply referred to as “schedule”). ) May be returned to the inquiry source. Here, taking the office environment as an example, the schedule can be assumed to be {meeting in conference room 1, meeting in conference room 2, and eating}. The schedule table system 211-1 may be provided inside the position information estimation server 2b.

  The time server 211-2 includes a clock device that detects the current time, and is a server device that outputs the detected current time to the position information estimation server 2b. Specifically, the time server 211-2 is an information processing device such as a personal computer that operates according to a program. Device.

  Further, the time server 211-2 has a simple database that inputs the current time and outputs the time zone, and can output the time zone to which the current time belongs in response to a request from the location information estimation server 2b. Good. Here, taking the office environment as an example, the time zone can be assumed to be {in the morning, lunch break, afternoon, overtime}. The time server 211-2 may be provided inside the position information estimation server 2b.

  In this embodiment, the case of the sensor system 211 that outputs sensor information, the schedule table system 211-1, and the time server 211-2 will be described as the sensor system, but the number of sensor systems may be any number. Another sensor system may be used.

  In addition to the sensor information used in the second embodiment, the sensor information acquisition unit 21a displays an observable variable (schedule and time zone in this example) correlated with a non-observable variable to be estimated. Collected from the system 211-1 and the time server 211-2, the collected observable variables are transferred to the non-observable variable estimation unit 23a.

  The parameter storage unit 22a uses the non-observable variables (user position and user) based on the sensor information and the observable variables (in this example, the schedule table and the time zone) that are correlated with the non-observable variables to be estimated. The parameters used for deriving the terminal position) are stored.

  The parameters may be stored in the parameter storage unit 22a in advance, or the parameters obtained by learning by inputting actual data may be stored in the parameter storage unit 22a as needed.

Further, the parameter storage unit 22 a stores the misplacement probability p _{drop in the same} manner as the parameter storage unit 22. The misplacement probability p _drop is not a constant, and may depend on the user position π _{t−1, the} user position π _t, or an observable variable.

  The output symbol storage unit 24a stores observable variables (schedules and time zones in this example) in chronological order in addition to sensor information that is an output symbol in the HMM described in the second embodiment. Further, the output symbol storage unit 24a may store a past estimation result (past user position or user edge position) used for estimation of the user position in time series or may be used for parameter learning. Data may be stored.

  The non-observable variable estimation unit 23a includes the sensor information and schedule and time zone collected by the sensor information acquisition unit 21a, the parameters stored in the parameter storage unit 22a, the previously derived user position and user terminal. Based on the position, the probability (a posteriori probability) that the user 1 and the user terminal 11 are in each state (position) when the sequence of sensor information is observed is obtained using the forward probability for the HMM. The probabilities obtained here are the position of the user 1 and the position of the user terminal 11 derived as random variables.

  FIG. 8 shows the user position and sensor device position (user terminal position) using other observable variables (for example, other sensor information, schedule, time zone variables, etc.) in addition to the sensor information. It is explanatory drawing explaining the probability model for estimating.

In FIG. 8, in addition to the model shown in FIG. 1, there are a plurality of sensor information x ₁ to x _n which are observable variables.

For example, in the example shown in FIG. 8, it is assumed that the user position π _{t at} time t has a correlation with the immediately preceding position π _t−1, and the conditional probability P (π _t | π _t−1 , x _1t ,. ·, X _nt ). It should be noted that parameters used for deriving non-observable variables that are required by adding a new sensor system are stored in advance in the parameter storage unit 22a.

  FIG. 9 is an explanatory diagram showing an example of parameters in the HMM. In FIG. 9, in addition to the parameters shown in FIG. 3, schedule table and time zone variables are introduced as observable variables correlated with the user position, and the existence probability by schedule and existence by time zone are parameters as parameters in the HMM. An observable variable output probability such as a probability may be included.

For example, scheduled by the existence probability refers to the probability that the user is in a position [pi _t in plan y _t at time t, the conditional probability P | represented by (π _t y _t).

Further, the existence probability by time zone indicates the probability that the user is at the position π t in the time zone z _t at time _t, and is indicated by a conditional probability P (π _t | z _t ).

  As shown in FIG. 9, when the state transition probability includes probabilities corresponding to a plurality of observable variables, the non-observable variable estimation unit 23a may integrate and use the probabilities for the observable variables.

For example, the non-observable variable estimation unit 23a integrates the inter-state movement probability, the schedule-specific existence probability, and the time-zone existence probability with respect to the position at the previous time, the schedule at the current time, and the time zone to which the current time belongs. A conditional probability P (π _t | π _t−1 , y _t , z _t ) may be used.

  Further, the non-observable variable estimation unit 23a indicates the state transition probability using the probability for each observable variable and the probability weights α, β, 1-α-β for each observable variable as follows: You may obtain | require by calculating such a weighted linear sum.

Based on the above, the non-observable variable estimation unit 23a uses other observable variables (for example, other sensor information, schedule, time zone variables, etc.) as output symbols in addition to the sensor information. The probability P _{i, k} (t) that the value of the person position π _t is the state s _i and the value of the sensor device position λ _t is the state s _k can be obtained.

Specifically, the non-observable variable estimation unit 23a first determines in which state the user 1 and the user terminal 11 are at the time 0 based on the initial existence probability. At this time, it is assumed that the user position π ₁ and the sensor device position λ ₁ are the same.

After that, the non-observable variable estimation unit 23a calculates the state transition probability, sensor device existence probability and sensor output probability of the HMM, sensor information and schedule and time zone at the current time, and past posterior probabilities at time t. A forward probability f _{i, k} (t) in which the user position is in the state s _i and the sensor device position is in the state s _k is _obtained by the following equation (Equation 15).

  Here, i represents in which state the user is in time t, and k represents in which state the sensor device is in. Further, m and l are symbols given for convenience in order to distinguish from i and k at time t, and represent the state of the user position and the position of the sensor device at time t-1.

Next, the non-observable variable estimation unit 23a uses the calculated forward probability f _{i, k} (t), and when the sensor information sequence is observed at time t, the user position is in the state s _i and the sensor posterior probability P _i that the device indicates the probability that the state s _k, a _k (t), and the posterior probability P _i (t) indicating the probability that a state s _i is the user position, the sensor device to each state s _k The posterior probability P _k (t) indicating the probability of being present is calculated using the following equation (Equation 16). In a few 16, P _i, k a (t) p _i, denoted as k (t), represents P _i a (t) p _i (t) and, P _k a (t) p _k (t) and Represents.

  By using such a method, the non-observable variable estimation unit 23a can derive the user position and the sensor device position in the form of random variables.

  FIG. 10 is a flowchart showing an example of position estimation processing of the position information estimation system according to the third embodiment of the present invention. In FIG. 10, the same operations as those shown in FIG. 6 are denoted by the same reference numerals.

  When there is an inquiry about the user position from the service providing server 3 (step S31), the sensor information acquisition unit 21a of the position information estimation server 2b correlates with the non-observable variable to be estimated in addition to the sensor information. Variables (plan and time zone in this example) are collected (steps S22 and S23).

  Subsequent processing (steps S24 to S26) is substantially the same as in the second embodiment.

  As described above, according to the present embodiment, it is possible to estimate the position of the user and the position of the user terminal with higher accuracy using a plurality of sensor systems.

(Fourth embodiment)
Next, an example of the position information estimation system according to the fourth embodiment of the present invention that has a plurality of sensor systems and a plurality of position information estimation servers and estimates the user position and the position of the user terminal will be described.

  FIG. 11 is a configuration diagram of a position information estimation system according to the fourth embodiment of the present invention. In addition to the configuration of the second embodiment, the position information estimation system of this embodiment includes other sensor information, a schedule table, and sensor systems 211-1 to 211-n that output variables such as time zones, Corresponding position information estimation servers 2-1 to 2-n are provided.

  FIG. 12 is a flowchart showing an example of position estimation processing of the position information estimation system in the fourth embodiment of the present invention. In FIG. 12, the same operations as those shown in FIG. 6 are denoted by the same reference numerals.

  The service providing server 3 inquires of the position information estimation servers 2 and 2-1 to 2-n about the user position, the position of the user terminal, and the incidental probability (step S31).

  The operations of the respective position information estimation servers 2 and 2-1 to 2-n are the same as those in the first embodiment.

  The service providing server 3 receives the user position, the user terminal position, and the incidental probability from the position information estimation servers 2 and 2-1 to 2-n, respectively, for example, from the position information estimation server by the sensor system having a high incidental probability. To provide services by trusting the user's location, or to derive a new user location by integrating the user location excluding the user location from the location information estimation server by the sensor system with low incidental probability May be.

  As described above, according to the present embodiment, when performing position estimation using a plurality of sensor systems, positions provided by incidental probabilities, such as not displaying user positions estimated by sensors with low incidental probabilities, etc. Information can be changed.

  In addition, when multiple sensor systems are used, the effectiveness of each sensor device can be calculated from the incidental probability, so the user position is estimated using only sensor information from the sensor device with high effectiveness. It becomes possible to do.

  According to each said embodiment, there exist the following effects.

  The first effect is that the user position can be estimated more accurately in consideration of misplacement of the sensor device.

  The reason is that the user's position is estimated based on a probabilistic model that also considers the possibility that the user and the sensor device are not in the same place, such as when the user misplaces the sensor device.

  As a second effect, it is possible to obtain the ratio (incident probability) that the user carries the sensor device.

  The reason is that by using a probability model in which the user position and the sensor device position can take different states, it becomes possible to obtain the posterior probability (incident probability) that the user position and the sensor device position match. Because.

  The third effect is that a single sensor can derive the probability that the user is carrying (or not carrying) the sensor device.

  The reason is that it is possible to determine the incidental probability by using a probability model that also considers the possibility that the user and the sensor device are not in the same place.

  The fourth effect is that even when the user does not carry the sensor device, the position where the user exists can be obtained with probability.

  The reason is that, when obtaining the user position, a probability model (for example, a hidden Markov model or a dynamic Bayesian network) derived based on a past state is used.

  The fifth effect is that the effectiveness of the sensor device can be calculated when a plurality of sensors are used.

  This is because the sensor information of a sensor with a high incidental probability is considered to be reliable.

  The sixth effect is that the presence presentation method can be changed depending on the incidental probability, such as not displaying the user position estimated by the sensor having a low incidental probability.

  This is because the incidental probability can be derived.

  In each embodiment described above, the illustrated configuration is merely an example, and the present invention is not limited to the configuration.

For example, the misplacement probability may be determined in relation to time of day or internal state or observable variables.
[Industrial applicability]
The present invention is applicable to a service providing system that provides a service using a user position.

It is a block diagram which shows the structural example of the positional infomation estimation system by the 1st Embodiment of this invention. It is explanatory drawing explaining the probability model for estimating the positional information by the 1st Embodiment of this invention. It is explanatory drawing which shows an example of the parameter in HMM by the 1st Embodiment of this invention. It is a flowchart which shows an example of the position estimation process by the position information estimation system by the 1st Embodiment of this invention. It is a block diagram which shows the structural example of the positional infomation estimation system by the 2nd Embodiment of this invention. It is a flowchart which shows an example of the position estimation process by the position information estimation system by the 2nd Embodiment of this invention. It is a block diagram which shows the structural example of the positional infomation estimation system by the 3rd Embodiment of this invention. It is explanatory drawing explaining the probability model for estimating the positional information by the 3rd Embodiment of this invention. It is explanatory drawing which shows an example of the parameter in HMM by the 3rd Embodiment of this invention. It is a flowchart which shows an example of the position estimation process by the position information estimation system by the 3rd Embodiment of this invention. It is a block diagram which shows the structural example of the positional infomation estimation system by the 4th Embodiment of this invention. It is a flowchart which shows an example of the position estimation process by the position information estimation system by the 4th Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 User 11 User terminal 2 Position information estimation server 20 Position estimation part 21 Sensor information acquisition part 22 Parameter memory | storage part 23 Unobservable variable estimation part 24 Output symbol memory | storage part 25 Additional probability estimation part 3 Service provision server 211 Sensor system 211 -1 Schedule table 211-2 Time server

Claims (20)

A position information estimation device that estimates the position of the user based on sensor information correlated with the position of a sensor device carried by the user,
Sensor information acquisition means for acquiring the sensor information;
In the probability model that the user and the sensor device can be in different positions, the position of the user is expressed, based on the probability model and the sensor information acquired by the sensor information acquisition means, a position estimation means for estimating a location of the user, only including,
The position estimating means includes
A sensor output parameter indicating a correlation between the sensor information and the position of the sensor device, and a misplacement parameter indicating a correlation between the position of the sensor device and the position of the user, a probability model parameter corresponding to the probability model Parameter storage means for storing as
The sensor information acquired by the sensor information acquisition unit as a non-observable variable that is information that cannot be observed from the outside, and the probability model parameters stored in the parameter storage unit. And a non-observable variable deriving unit for deriving based on the position information estimating device. The sensor information acquisition means further acquires an observable variable having a correlation with the position of the user,
The parameter storage means further stores an observable variable output parameter indicating a correlation between the observable variable and the position of the user as the probability model parameter,
The non-observable variable deriving unit is configured to determine the position and use of the sensor device based on the sensor information and the observable variable acquired by the sensor information acquiring unit and the probability model parameter stored in the parameter storage unit. It derives the position of the person, positional information estimating apparatus according to claim 1. The position estimation means further comprises history storage means for storing the estimated position of the sensor device and the estimated position of the user according to a time series,
The non-observable variable deriving means includes at least the position of the sensor device and the position of the user stored in the history storage means, the sensor information acquired by the sensor information acquisition means, and the parameter storage means a probability model parameters, based on, derives the position and the current position of the user of the current of the sensor device, the position information estimating apparatus according to claim 1 or 2. The stochastic model is a hidden Markov model or a dynamic Bayesian network model that estimates the non-observable variable that changes with time based on at least the sensor information;
The parameter storage means uses, as the probability model parameter, a parameter for formulating the position of the sensor device and the position of the user at each time as the internal state of the hidden Markov model or the dynamic Bayesian network model. Remember,
The non-observable variable derivation means is based on at least the sensor information acquired by the sensor information acquisition means and the parameters stored in the parameter storage means, and the current position of the sensor device and the current use. The position information estimation apparatus according to any one of claims 1 to 3 , wherein a position of a person is derived. Sensor wherein the parameter storage means, as said parameter, indicating at least the probability of the sensor information x _t is output if the position of the sensor device is shifted to the internal state s _t corresponding to the time t at time t Storing an output probability and a misplacement probability that is a probability that the user misplaces the sensor device;
The non-observable variable deriving unit is configured to generate the sensor device and the sensor at the current time based on the sensor information acquired by the sensor information acquiring unit and the sensor output probability and the misplacement probability stored in the parameter storage unit. The position information estimation apparatus according to claim 4 , wherein the position of the current sensor device and the position of the current user are derived by obtaining a probability that each of the users is in each internal state. The position information estimation apparatus according to claim 5 , wherein the misplacement probability is determined in relation to time or the internal state, or the sensor information. The sensor information acquisition means acquires an observable variable having a correlation with the position of the user,
The position information estimation apparatus according to claim 6 , wherein the misplacement probability is determined in relation to time or the internal state, the sensor information, or the observable variable. The sensor information acquisition means acquires an observable variable y having a correlation with the position of the user,
The parameter storage means further includes the parameter,
First, the probability that the sensor device and the user are in the internal state s _i (i is a natural number, and when each internal state is expressed as internal states s ₁ to s _j , 1 <i ≦ j Initial existence probability indicating
The user is internal state s _t-1 from the state transition probability is the probability of transition to the internal state s _t (t is a natural number corresponding to the time of observation: 1 <t) and,
Stores, and observable variable output probability indicating the probability of the observable variable y _t is output when the sensor device is shifted to the internal state s _t,
The non-observable variable deriving unit is configured to detect the sensor device at the current time based on the initial existence probability, the state transition probability, the observable variable output probability, the sensor output probability, and the misplacement probability stored in the parameter storage unit. The position information estimation apparatus according to claim 5 or 6 , wherein the position of the current sensor device and the position of the current user are derived by obtaining a probability that each of the users is in each internal state. The non-observable variables deriving means, as the state transition probability, the internal state at the previous time s _t-1 and the conditional probability P for observability variable y _t at the current time _{(s t | s t-1} , y t) The position information estimation apparatus according to claim 8 , wherein: The non-observable variables deriving means, as said probability model, the hidden Markov model, upon the output symbols of the sensor information, sequence {x ₁ of the sensor information x _t at the current time t, x _2, .., X _t } are observed, the posterior probabilities indicating the probability that each of the sensor device and the user is in each internal state s _i are _expressed as conditional probabilities P (s _i | x ₁ , x ₂ , .., X _t ) according to any one of claims 4 to 9 , wherein the position information estimation device is obtained using a forward algorithm for the hidden Markov model. The non-observable variable deriving means has a random variable representing the position of the user at time t as π _t and a random variable representing the position of the sensor device as λ _t, and the position of the user and the sensor The position information estimation apparatus according to any one of claims 1 to 10 , wherein calculation is performed by regarding a set of device positions (π _t , λ _t ) as one variable. The parameter storage means stores a misplacement probability p _drop which is a probability that the user misplaces the sensor device,
The non-observable variable deriving means sets a random variable representing the position of the user at time t as π _t , a random variable representing the position of the sensor device as λ _t, and a conditional probability P related to the position of the sensor device. (Λ _t | λ _t-1 , π _t-1 , π _t )

The position information estimation apparatus according to any one of claims 1 to 11 . The position information estimation apparatus according to claim 12 , wherein the p _drop is not a constant but depends on the π _t-1 or the π _t or other sensor information. The sensor information acquisition means acquires an observable variable having a correlation with the position of the user,
The position information estimation apparatus according to claim 13 , wherein the p _drop is not a constant but depends on the π _t−1 or the π _t or other sensor information, or the observable variable. The probabilistic model is a hidden Markov model that estimates the non-observable variable that changes with time based on at least the sensor information;
The parameter storage means includes
A sensor output probability indicating a probability that the sensor information is output when the position of the sensor device transitions to an internal state of the hidden Markov model corresponding to the time t at time t;
First, the probability that the sensor device and the user are in the internal state s _i (i is a natural number, and when each internal state is expressed as the internal state s ₁ to s _j , 1 <i ≦ j is satisfied. ) Initial existence probability indicating
The user is internal state s _t-1 from the state transition probability is the probability of transition to the internal state s _t (t is a natural number corresponding to the time of observation: 1 <t) and stores,
The non-observable variable deriving means includes an initial existence probability I _i of each state s _i at time 0, a state transition probability, a sensor output probability, sensor information at the current time, and a calculation result of a forward probability at the previous time. At time t, the user obtains a forward probability f _{i, k} (t) in which the user is in the internal state s _i and the sensor device is in s _k by the following equation:

Further, using the determined forward probability f _{i, k} (t), the position of the user is in the state s _i when the sensor information sequence is observed at time t, and the sensor device is in the state s _k . The posterior probability P _{i, k} (t) indicating a certain probability, the posterior probability P _i (t) indicating the probability that the user's position is in the state s _i, and the probability that the sensor device is in each state s _k The posterior probability P _k (t) to be shown is calculated using the following equation:

The position information estimation apparatus according to claim 12 or 13 . The sensor information acquisition means acquires an observable variable y having a correlation with the position of the user,
The probability model is a hidden Markov model that estimates the non-observable variable that changes with time based on at least the sensor information and the observable variable y,
The parameter storage means includes
A sensor output probability indicating the probability that the sensor information is output if the position of the sensor device is shifted to the internal state s _t of the hidden Markov model at time t,
And observability variable output probability indicating the probability of the observable variable is output if the position of the user is shifted to the internal state s _t at time t,
First, the probability that the sensor device and the user are in the internal state s _i (i is a natural number, and when each internal state is expressed as the internal state s ₁ to s _j , 1 <i ≦ j is satisfied. ) Initial existence probability indicating
The user is internal state s _t-1 from the state transition probability is the probability of transition to the internal state s _t (t is a natural number corresponding to the time of observation: 1 <t) and stores,
The non-observable variable deriving means includes initial existence probability I _i of each state s _i at time 0, state transition probability, sensor output probability, observable variable output probability, sensor information x and observable variable y at the current time, And the forward probability f _{i, k} (t) where the user is at s _i and the sensor device is at s _k at time t from the calculation result of the forward probability at the previous time, and

Further, using the determined forward probability f _{i, k} (t), the position of the user is in the state s _i when the sequence of sensor information is observed at time t, and the sensor device is in the state s _k . A posteriori probability P _{i, k} (t) indicating the probability, a posteriori probability P _i (t) indicating the probability that the user's position is in the state s _i , and the sensor device is in each state s _k Find the posterior probability P _k (t) indicating the probability using the following equation:

The position information estimation apparatus according to claim 12 or 13 . The position estimation means derives an incidental probability representing a ratio of the user carrying the sensor device using the probability of the position of the user and the position of the sensor device estimated based on the probability model. The position information estimation apparatus according to claim 1, further comprising an auxiliary probability deriving unit. The incidental probability deriving means has π _t = λ _t when the user position probability π _t and the sensor device position probability λ _{t at} time _t are obtained by the non-observable variable derivation means. The position information estimation apparatus according to claim 17 , wherein the event posterior probability P _r (π _t = λ _t | x ₁ , x ₂ ,..., X _t ) is obtained. A position information estimation method performed by a position information estimation device that estimates the position of the user based on sensor information correlated with the position of a sensor device carried by the user,
A sensor information acquisition step for acquiring the sensor information;
Position estimation that represents the position of the user in the probability model that the user and the sensor device can be at different positions, and estimates the position of the user based on the probability model and the sensor information and the step, only including,
The position estimating step includes:
A sensor output parameter indicating a correlation between the sensor information and the position of the sensor device, and a misplacement parameter indicating a correlation between the position of the sensor device and the position of the user, a probability model parameter corresponding to the probability model As a parameter storage means,
A position information estimation method in which the position of the sensor device and the position of the user are derived based on the sensor information and the probability model parameters as non-observable variables that are information that cannot be observed from the outside . A position information estimation program for causing a computer to execute position information estimation processing for estimating the position of the user based on sensor information correlated with the position of a sensor device carried by the user,
Sensor information acquisition processing for acquiring the sensor information;
Position estimation that represents the position of the user in the probability model that the user and the sensor device can be at different positions, and estimates the position of the user based on the probability model and the sensor information processing and, only including,
In the position estimation process,
A sensor output parameter indicating a correlation between the sensor information and the position of the sensor device, and a misplacement parameter indicating a correlation between the position of the sensor device and the position of the user, a probability model parameter corresponding to the probability model As a parameter storage means,
Position information estimation processing for deriving the position of the sensor device and the position of the user as non-observable variables that are information that cannot be observed from the outside based on the sensor information and the probability model parameters , A position information estimation program to be executed by a computer.

Images