JP2010139325A - Position estimation apparatus and position estimation method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a position estimation apparatus and a position estimation method which can improve the accuracy of estimating the position of a mobile object. <P>SOLUTION: A motion parameter selection section 202 acquires the attribute information of the mobile object corresponding to the ID information of the mobile object and the attribute information of a floor corresponding to positional information from a database 112 and, on the basis of the acquired attribute information, selects an optimum motion parameter. A distribution update section 203, using the positional distribution parameter of the mobile object at the previous time held by a distribution parameter hold section 204, the positional information, and the selected motion parameter, calculates the positional distribution parameter of the mobile object at the present time. A most-probable position calculation section 205 calculates the most-probable position as the position of the mobile object from the calculated positional distribution parameter of the mobile object. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a position estimation device and a position estimation method for estimating the position of a moving object.

  There is a growing demand for security management systems that monitor the tracking and behavior of suspicious persons in office spaces and public spaces, and safety management systems that detect intruders in hazardous areas at factories, construction sites, disaster sites, and the like. In realizing such a system, it is important to estimate the position of the monitoring target (hereinafter referred to as “observation target”) with high accuracy.

  Estimating the position of an observation target with high accuracy means that even if the observation target moves from moment to moment and the surrounding conditions of the observation subject change in various ways as a result of movement, the position estimation accuracy is not significantly degraded. This refers to continuously estimating the position of the observation target.

  The response to the surrounding situation of the observation target is calculated by calculating the probability density distribution of the object position from multiple modal information about the observation target, combining these probability density distributions as a weighted linear sum, and observing from the obtained likelihood distribution A method for obtaining the position of an object has been proposed. In addition, a method for improving the accuracy of position estimation of an observation target using past observation data has been proposed (see, for example, Patent Document 1).

  Here, Patent Document 1 describes a position estimation method using a likelihood distribution, which will be described below. Patent Document 1 discloses a technique for estimating the position of a moving object even when the accuracy of a sensor temporarily deteriorates due to an external factor or an observation target cannot be observed in a method using a plurality of sensors with different accuracy. ing. In particular, Patent Document 1 is characterized in that the probability density distribution obtained from the observation result and the probability density distribution at the previous time are integrated as a weighted linear sum to obtain the probability density distribution at the current time.

  That is, in Patent Document 1, the probability density distribution based on the observation result is not simply integrated, but is weighted linearly summed according to the reliability of the sensor. Thus, in Patent Document 1, even when there is a sensor whose observation accuracy is deteriorated due to an external factor, position estimation can be continuously performed without significantly degrading the observation accuracy.

  In the position estimation method described in Patent Document 1, position estimation is performed using Bayes' theorem, but position estimation accuracy is degraded due to the movement of the object. In other words, it is not sufficient to use the Bayes' theorem to degrade the observation accuracy when the position of a moving object is estimated. Incidentally, Bayes' basic theorem is shown below.

P (X | Y) ∝P (X) × P (X | Y)
In Bayesian theorem, P (X) is the certainty of X before event Y occurs, P (X | Y) is the certainty of X after event Y occurs, and P (Y | X) is X The conditional probability that event Y will occur under the generated conditions is shown. In Bayes, it is defined as the likelihood (likelihood) of X when event Y occurs. In application to position estimation, the parameter of distribution (position) is estimated by increasing the number of events and generalizing as a continuous function. However, it is assumed here that the distribution is normal.

f (μ) ∝ g (X | μ) x f (μ)
In the above equation, g (x | μ) is the probability density when the position is X, and represents the likelihood (likelihood) that the position is X. The movement of the observation target results in a state in which the variance of the observation error (likelihood distribution) spreads as a result, and the likelihood distribution in such a state is multiplied by the prior distribution, so that the position estimation accuracy deteriorates.
Japanese Patent Laid-Open No. 2005-141687

  However, when the likelihood distribution is simply applied to estimate the position of a moving object, there is a problem that if the past observation data is used to improve the accuracy of position estimation, the position estimation accuracy deteriorates. is doing. In contrast to this accuracy degradation, extreme accuracy degradation can be suppressed by widening the variance of the likelihood distribution, but the accuracy of the observation data is lowered, resulting in degradation of position estimation accuracy. For this reason, when the past observation data is used in the position estimation of the moving object, it is difficult to improve the position estimation accuracy compared to the position estimation of the stationary object.

  The present invention has been made in view of such points, and an object thereof is to provide a position estimation device and a position estimation method that improve the position estimation accuracy of a moving object.

  The position estimation apparatus of the present invention includes a motion parameter selection unit that selects a motion parameter including an average of a moving speed of a moving object and a variance of the moving speed, and different points in time of the moving object using the selected motion parameter. A distribution updating unit for calculating a probability distribution of the position of the moving object from a plurality of observation data in the method, and a maximum likelihood for calculating a position having the highest probability as the position of the moving object from the calculated probability distribution of the position of the moving object And a position calculation unit.

  The position estimation method of the present invention includes a motion parameter selection step of selecting a motion parameter including an average of a moving speed of a moving object and a variance of the moving speed, and different time points of the moving object using the selected motion parameter. A distribution update step of calculating a probability distribution of the position of the moving object from a plurality of observation data in the method, and a maximum likelihood calculating a position having the highest probability as the position of the moving object from the calculated probability distribution of the position of the moving object A position calculating step.

  According to the present invention, it is possible to improve the position estimation accuracy of a moving object.

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

(Embodiment 1)
FIG. 1 is a block diagram showing a configuration of a position estimation system according to Embodiment 1 of the present invention. In FIG. 1, the position estimation system includes a plurality of detection units 101, a communication unit 111, a database unit 112, a position estimation unit 113, and a position display unit 114 installed on the floor.

  One or more detection units 101 are installed on the floor, observe a moving object existing on the floor, and include a tag ID (hereinafter referred to as “ID information”) attached to the moving object and a position of the moving object. Detect information. The moving body information including the detected ID information and position information is output to the communication unit 111 of the position estimation device 110.

  The communication unit 111 is connected to the plurality of detection units 101 directly or via a network, and outputs moving body information including ID information and position information output from each detection unit 101 to the position estimation unit 113.

  The database unit 112 manages moving object attribute information and floor attribute information. The attribute information of the moving object and the floor attribute information will be described later.

  The position estimation unit 113 estimates the position of the moving object based on the ID information and position information output from the detection unit 101 and the attribute information stored in the database unit 112, and the estimated position information is displayed in the position display unit 114. Output to.

  Based on the position information output from the position estimation unit 113, the position display unit 114 displays the movement trajectory (flow line) of the moving object on the monitor in real time. Note that the position information may be displayed on the monitor by a method other than the flow line. For example, the position display unit 114 may display a message on the monitor when a moving object enters a specific area. Further, the position display unit 114 may be presented via an external output device (for example, a printer) other than the monitor. For example, the position display unit 114 may sound an alarm sound when a moving object enters a specific area.

  Such a position estimation system is used for grasping (monitoring) a suspicious person's position and a suspicious person's movement history as a security system in an office or the like. In addition, the position estimation system is used for flow analysis of people and objects in public facilities and exhibition halls.

  FIG. 2 is a block diagram showing an internal configuration of the position estimation unit 113 shown in FIG. In FIG. 2, the position estimation unit 113 includes an information separation unit 201, a motion parameter selection unit 202, a distribution update unit 203, a distribution parameter holding unit 204, and a maximum likelihood position calculation unit 205.

  The information separation unit 201 acquires the ID information and position information output from the detection unit 101 via the communication unit 111, outputs the acquired ID information and position information to the motion parameter selection unit 202, and updates the distribution of the position information. The data is output to the unit 203.

  The movement parameter selection unit 202 separates the ID information and the position information output from the information separation unit 201, and the database unit stores the attribute information of the moving object corresponding to the ID information and the attribute information of the floor corresponding to the position information. 112. The exercise parameter selection unit 202 selects an optimal exercise parameter based on the acquired attribute information, and outputs it to the distribution update unit 203. The motion parameter is a parameter representing what kind of motion the moving object is performing, and here means the average of the moving speed of the moving object and the variance of the moving speed. The ID information separated here does not need to be a final ID for identifying a moving object, and can be any value as long as it is information that can be uniquely converted into a final ID. Good.

  The distribution updating unit 203 outputs from the distribution parameter holding unit 204 distribution parameters (average and variance of the moving object position distribution) representing the probability distribution of the moving object position at the previous time, and the information separating unit 201. The obtained position information (coordinate position) and the motion parameter (average and variance of the velocity distribution of the moving object) output from the motion parameter selection unit 202 are acquired. The distribution updating unit 203 obtains a distribution parameter representing the probability distribution of the position of the moving object at the current time based on the acquired distribution parameter representing the probability distribution of the position of the moving object, the position information, and the motion parameter, and a distribution parameter holding unit To 204. That is, the distribution updating unit 203 updates the average and variance of the moving object position distribution based on the normal distribution. It is assumed that the distribution parameter of the position of the moving object at time t = 0 is stored in advance in the distribution parameter holding unit 204 as an initial value.

  The distribution parameter holding unit 204 holds the distribution parameter of the position of the moving object, and outputs it to the distribution update unit 203 and the maximum likelihood position calculation unit 205. The distribution parameter holding unit 204 receives the output of the distribution updating unit 203 and replaces the held distribution parameters (average and variance of the distribution of the position of the moving object). In the initial state, the initial value of the distribution parameter is held.

  The maximum likelihood position calculation unit 205 acquires the distribution parameter of the position of the moving object from the distribution parameter holding unit 204, obtains the position (coordinate position) with the highest probability as the position of the moving object, calculates the position as the maximum likelihood position, The data is output to the display unit 114.

  The description from here is about the method of selecting the exercise parameter of the exercise parameter selection unit 202. A plurality of models to be observed are set for each parameter value, determined in advance for each model, and stored in the motion parameter selection unit 202 as a conversion table.

  There are mainly the following three methods for selecting motion parameters. One is a selection method based on the attribute information of the moving object, two is a selection method based on the attribute information of the floor on which the moving object moves, and three is based on the movement history of the moving object. It is a method to select. Hereinafter, each of these three selection methods will be described.

  The method described first is a method of selecting a motion parameter based on attribute information of a moving object. FIG. 3 is a diagram illustrating an example of attribute information of a moving object. Here, it is assumed that the moving object is a person. As shown in FIG. 3, the attribute information of the moving object is obtained by associating the sex, age, and classification with the ID information. For the classification, for example, a value obtained by a general statistical method (for example, clustering analysis, principal component analysis, etc.) is used.

  The motion parameter selection unit 202 acquires ID information from the information separation unit 201, searches the attribute information of the moving object stored in the database unit 112 based on the acquired ID information, and acquires corresponding attribute information.

  The motion parameter selection unit 202 acquires the average speed of the moving object and the variance of the speed from the acquired attribute information using a conversion table held in advance. The motion parameter selection unit 202 updates the motion parameter of the moving object at the current time using the acquired value, and outputs it to the distribution update unit 203. As shown in FIG. 4, the conversion table is a two-dimensional table with gender and age as axes, and motion parameters are set for individual elements. For example, an elderly person has a low average moving speed (slow moving speed) and a narrow variance. In addition, for children, the average moving speed is fast and the variance is wide. In the conversion table, such values are set in advance based on rules of thumb and history.

  Next, a method for selecting motion parameters based on the attribute information of the floor on which the moving object moves will be described. FIG. 5 is a diagram illustrating an example of floor attribute information. Here, the floor is divided into fixed sections (each block), and the areas in the floor are specified by the floor area information for identifying the divided individual areas. As shown in FIG. 5, the floor attribute information is obtained by associating floor area information with average residence time, average of moving speed, floor type (classification here), and distribution of moving speed. The average residence time indicates the average residence time of a moving object that has passed through this region in the past. The average moving speed indicates the average passing speed of a moving object that has passed through this area in the past. The floor type is information representing the characteristics of this area, and is represented by the classification shown in FIG. 3 as an example here. In this case, the floor type is set with the most classification among all moving objects that have passed through this area or stayed for a long time. Further, the dispersion of the moving speed indicates the dispersion of the passing speed of the moving object that has passed through this region.

  The exercise parameter selection unit 202 acquires the position information from the information separation unit 201, and searches the floor area information stored in the database unit 112 based on the acquired position information to acquire the corresponding floor attribute information.

  The motion parameter selection unit 202 updates the motion parameter of the moving object at the current time based on the acquired average and variance of the moving speed, and outputs the updated motion parameter to the distribution update unit 203.

  The motion parameter selection unit 202 may not match the classification of the moving object acquired from the database unit 112 based on the ID information acquired simultaneously with the average and variance of the moving speed and the floor type. At this time, the motion parameter selection unit 202 applies the motion parameter corresponding to the classification or the default motion parameter held in advance by the position estimation unit 112. Accordingly, the motion parameter selection unit 202 can apply appropriate motion parameters to a moving object whose motion pattern can be predicted in advance or a moving object that is expected to deviate greatly from the variance.

  Next, a method for selecting a motion parameter based on the movement history of a moving object will be described. FIG. 6 is a diagram illustrating an example of a movement history of a moving object. As shown in FIG. 6, as the movement history of the moving object, floor position information, a movement destination, and accuracy are associated with ID information. The ID information indicates a moving object that has passed through this area in the past. The floor position information is information for identifying a specific position within the floor. Further, the movement destination indicates the speed and moving direction at which the moving object has passed a specific position in the floor. Further, the accuracy indicates the number of times the moving object has moved at the speed indicated by the destination when the number of times that the moving object has passed the specific position on the floor is 100.

  The movement parameter selection unit 202 searches the floor position information of the movement history of the moving object stored in the database unit 112 based on the position information acquired from the information separation unit 201, and calculates the probability distribution of the movement destination calculated from the history. Get the parameter that represents. The motion parameter selection unit 202 updates the current average speed and variance as motion parameters from the acquired parameters, and outputs them to the distribution update unit 203.

  Note that the movement history of a moving object may be the average of a plurality of moving objects in the past, instead of information on a specific moving object.

  The exercise parameter selection unit 202 may select which of the above three selection methods is used in advance depending on the use case of the position estimation method. For example, when there is no floor restriction on movement (for example, the traveling direction or traveling speed of a staircase or a moving sidewalk), it is conceivable to select a motion parameter based on the attribute of the moving object. Further, when there is a history regarding the movement speed (staying time, stagnation time) of a person at an exhibition or the like, it is conceivable to select an exercise parameter based on the history in preference to attribute information or floor information. In addition, the motion parameter selection unit 202 may dynamically switch between the above three selection methods. In this case, there is a method of setting a priority order to apply the selection method for each place or moving object. Incidentally, the motion parameter selection unit 202 may combine the above three methods.

  The following description is about moving object position estimation processing. Here, the method for selecting the first motion parameter will be described as an example with reference to FIGS.

  The distribution update unit 203 is a parameter related to the distribution of the position of the moving object at the previous time from the distribution parameter holding unit 204, the current observation value of the moving object from the information separation unit 201, and the motion parameter from the motion parameter selection unit 202. And get. Subsequently, the distribution updating unit 203 calculates parameters representing the position distribution of the moving object at the current time using them, and outputs them to the distribution parameter holding unit 204.

The distribution updating unit 203 acquires the average speed and the variance of the speed of the moving object, which are the motion parameters, from the motion parameter selection unit 202. Further, the distribution updating unit 203 acquires, from the distribution parameter holding unit 204, the average μ _t of the moving object position and the variance λ _t ² as the distribution parameters of the position of the target moving object at time t = t.

It is assumed that the distribution parameter of the position of the moving object at time t = 0 is stored in advance in the distribution parameter holding unit 204 as an initial value. The initial values of the distribution parameters are preferably μ _t = 0 and λ ₀ ² = + ∞.

At this time, the distribution updating unit 203 acquires the observation value y _{t + 1} from the information separating unit 201 at time t = t + 1. Next, the distribution updating unit 203 calculates the average μ _{t + 1} and the variance λ _{t + 1} ² of the moving object positions as the parameters of the distribution of the moving object position x _{t + 1} at time t = t + 1 by the following equation (1).

However, σ ² is the variance in the observation model, and v _t and τ _t ² are the average and variance of the moving speed of the object at time t selected by the motion parameter selection unit 202. (Note that the observed values are observed according to the distribution of mean 0 and variance σ ² around the true value)

The distribution updating unit 203 stores the average μ _{t + 1} of the moving object positions and the variance λ _{t + 1} ² obtained by the above equation in the distribution parameter holding unit 204.

The maximum likelihood position calculation unit 205 acquires the distribution parameter of the position of the moving object from the distribution parameter holding unit 204, obtains the position with the highest probability as the position of the moving object, calculates the position as the maximum likelihood position, and outputs the position to the position display unit 114. Output. Specifically, the maximum likelihood position calculation unit 205 acquires the average μ _{t + 1} of the distribution of the position of the moving object at time t = _{t + 1} and the variance λ _{t + 1} ² that are parameters stored in the distribution parameter storage unit 204. Next, the maximum likelihood position calculation unit 205 calculates μ _{t + 1} as the maximum likelihood value of the position of the moving object at time t = t + 1, and outputs the calculated maximum likelihood position μ _{t + 1} to the position display unit 114.

  Hereinafter, in a simulation example of the position of a person walking in a hurry, an example of position estimation using a motion model selected using an ID parameter will be shown. FIG. 4 shows the average value and standard deviation of the speed when a person walks in a hurry, classified by gender and age, and is stored in the database unit 112 as an exercise parameter.

  Assume that the true position of a 60-year-old male at time t = t is obtained as shown in FIG. 7 (FIG. 7 is a simulation result of 1000 samples). At this time, according to the table of FIG. 4, the moving speed of a 60-year-old male is 1.92 (m / s) on average, and the standard deviation is 0.21, and the true position at time t = t + 1 is calculated by simulation. For example, it is obtained as shown in FIG.

  Assuming a model in which observed values are distributed around a true value according to a normal distribution with mean 0 and variance 636, the observed value at this time is obtained as shown in FIG. 9, for example. At this time, based on the moving speed model of a 60-year-old male, the distribution of the maximum likelihood value at the position at time t = t + 1 obtained from the above update formula based on each observation value is as shown in FIG.

  The error distribution obtained by comparing the maximum likelihood value obtained by this embodiment with the true position corresponding to this maximum likelihood value is as shown in FIG. 11, and the average is 16.1.

  On the other hand, based on a general moving speed model according to the prior art, the distribution of maximum likelihood values at time t = t + 1 obtained from the above update formula based on each observation value is as shown in FIG. The distribution is as shown in FIG. 13 and the average is 18.4. It is assumed that a general moving speed model has an average of 2.24 m / s and a standard deviation of 0.69.

  As described above, according to the present embodiment, a motion parameter is selected based on ID information of a moving object, and a moving object is obtained from a plurality of observation data at different time points of the moving object using the selected motion parameter. It is possible to estimate the position of the moving object with high accuracy by calculating the likelihood distribution of the position of, and calculating the position of the moving object from the calculated likelihood distribution.

  The position estimation device and the position estimation method according to the present invention can be applied to a security management system, a safety management system, and the like.

The block diagram which shows the structure of the position estimation system which concerns on Embodiment 1 of this invention. The block diagram which shows the internal structure of the position estimation part shown in FIG. The figure which shows an example of the attribute information of a moving object The figure which shows an example of a conversion table The figure which shows an example of the attribute information of a floor The figure which shows an example of the movement history of a moving object The figure which shows the simulation result of the true position in the time t = t The figure which shows the simulation result of the true position in the time t = t + 1 The figure which shows the simulation result of the observed value of the position in time t = t + 1 The figure which shows distribution of the maximum likelihood value of the position in time t = t + 1 The figure which shows distribution of the error of the maximum likelihood value of a position in time t = t + 1, and a true position The figure which shows distribution of the maximum likelihood value of the position in time t = t + 1 based on a general moving speed model The figure which shows distribution of the error of the maximum likelihood value of a position in time t = t + 1 and a true position based on a general moving speed model

Explanation of symbols

DESCRIPTION OF SYMBOLS 101 Detection part 111 Communication part 112 Database part 113 Position estimation part 114 Position display part 201 Information separation part 202 Motion parameter selection part 203 Distribution update part 204 Distribution parameter holding part 205 Maximum likelihood position calculation part

Claims (5)

A motion parameter selector for selecting motion parameters including an average of moving speeds of moving objects and dispersion of moving speeds;
A distribution updating unit that calculates a probability distribution of the position of the moving object from a plurality of observation data at different time points of the moving object, using the selected motion parameter;
A maximum likelihood position calculation unit that calculates a position having the highest probability as the position of the moving object from the calculated probability distribution of the position of the moving object;
A position estimation apparatus comprising:   The position estimation device according to claim 1, wherein the motion parameter selection unit selects a motion parameter based on an attribute of the moving object.   The position estimation device according to claim 1, wherein the motion parameter selection unit selects a motion parameter based on an attribute of a place where the moving object moves.   The position estimation device according to claim 1, wherein the motion parameter selection unit selects a motion parameter based on a movement history of the moving object. A motion parameter selection step for selecting motion parameters including an average of the moving speed of the moving object and a variance of the moving speed;
A distribution update step of calculating a probability distribution of the position of the moving object from a plurality of observation data at different time points of the moving object using the selected motion parameter;
A maximum likelihood position calculating step of calculating a position having the highest probability as the position of the moving object from the calculated probability distribution of the position of the moving object;
A position estimation method comprising:

Images