JP2013053930A - System, method, and program for locating article - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an article locating system capable of accurately identifying the location of an indoor target article with a reduced amount of computation.SOLUTION: RSSI values of test radio wave signals from test transmitters PL1 to PLm for pre-learning, disposed at known locations, are acquired by receivers AP1 to APn disposed at specific locations successively for a predetermined period of time. RSSI values of test radio wave signals from locations without test transmitters are estimated by applying Gaussian process regression on the acquired RSSI values, which are used by a pre-learnt RSSI value map generator 12 to generate a pre-learnt RSSI value map covering the whole indoor space. RSSI values of a radio wave signal from a target article T to be located are then acquired by the receivers AP1 to APn as target article RSSI values, and the location of the target article T is estimated by a target article location estimator 14 on the basis of the acquired target article RSSI values and the RSSI value map.

Description

  The present invention relates to an article position specifying system, an article position specifying method, and an article position specifying program, and in particular, for each article using a radio wave transmitter (for example, a wireless tag, a wireless device, etc.) attached to an article placed indoors. The present invention relates to an article position specifying system, an article position specifying method, and an article position specifying program for specifying a position.

  In recent years, where indoors are used to manage various equipment assets, and to manage various sensors installed for the purpose of "energy visualization system", etc. The position identification mechanism that can easily detect the movement has become an important issue.

  Conventionally, as a technology for specifying the position of an article using a radio wave signal indoors, conventionally, an RSSI value (Received Signal Strength Indicator) of a radio wave signal emitted from a target article whose position is to be determined is measured, and the radio wave is transmitted. A technique for estimating the position of a target article that is a source of signals has been used.

  Here, as a conventional technique for estimating the position of an article that is a source of a radio wave signal based on the measured RSSI value, as shown in FIG. 8, each receiver (access point: A method of simultaneously measuring RSSI values of radio wave signals received by (Access Point) and estimating a distance by applying a propagation loss equation applied according to the frequency band of the received radio wave signals and the indoor type thereof. It is used. FIG. 8 is a schematic diagram for explaining a conventional technique for specifying the position of an article from the RSSI value of a radio wave signal.

  That is, in FIG. 8, the target article T whose position is to be specified transmits a radio wave signal of a specific frequency band. In addition, a plurality of receivers AP1 to APn (n = 5 in the case of FIG. 8) whose arrangement locations are known in advance receive the radio wave signal from the target article T at the same time, and each of the received radios The RSSI value of the radio signal is transmitted to a location estimation device LS (Location Estimating System).

  Here, the position estimation device LS is wireless in each of a plurality of receivers AP1 to APn whose locations are known in advance in order to specify the position of the target article T that is the source of the radio wave signal. The coordinate position of the target article T is estimated by applying a propagation loss formula uniquely determined by the type, frequency band, and the like of the used radio wave signal to the result of collecting the RSSI values of the radio signal.

  However, when applying the conventional method for estimating the position of an article as shown in FIG. 8, it is necessary that the propagation loss equation and its coefficient be clarified in advance. Furthermore, in an actual indoor environment, the indoor radio wave environment is not uniform, and the target article T is located in the vicinity of furniture or walls such as desks and article storage boxes that are present indoors. And the case of being located in a wide space near the center, the characteristics of the propagation loss regarding the radio wave signal received by each of the receivers AP1 to APn are different. There is a problem that an error is likely to occur in the estimated article position only by the estimation result of the article position estimated using only one propagation loss equation uniquely determined by the above.

  On the other hand, in order to cope with different indoor radio wave environment conditions, it is very difficult to predetermine a formula representing each propagation loss for every indoor location, which is not practical.

  As a method for solving such problems, for example, a prior learning method as described in “RADAR: An In-building RF-based User Location and Tracking System” by P. Bahl et al. The technique for specifying the position of the article is now being used. That is, as shown in FIG. 9, using test transmitters PL1 to PLm (m = 9 in the case of FIG. 9) prepared for RSSI value measurement for pre-learning, different radio environment conditions for pre-learning Pre-learning data is obtained by measuring in advance RSSI values of test radio wave signals when transmitting from a plurality of arbitrarily determined indoor locations at each receiver AP1 to APn (n = 5 in FIG. 9). Save it as a database.

  Here, FIG. 9 is a schematic diagram for explaining the estimation method of the RSSI value of the radio wave signal by the conventional pre-learning method, and FIG. 9A is the RSSI of the test radio wave signal in the pre-learning phase. FIG. 9B shows a situation in which the position of an actual target article whose position is to be specified is estimated in the position estimation phase after the prior learning.

  In other words, in the pre-learning phase in FIG. 9A, the points for pre-learning are prepared using test transmitters PL1 to PLm (m = 9 in the case of FIG. 9) prepared in advance for collecting pre-learning data. As a position estimation device, a test radio wave signal is transmitted at each of a plurality of indoor locations arbitrarily determined in advance and measured at each receiver AP1 to APn (n = 5 in FIG. 9). It transmits to LS and preserve | saves beforehand as data for prior learning in the database of the position estimation apparatus LS. Here, each place where each of the test transmitters PL1 to PLm is arranged is a position that is known in advance, and each of the receivers AP1 to APn that receives the test radio wave signals from the test transmitters PL1 to PLm. The location is also known in advance.

  Thereafter, in the position estimation phase of FIG. 9B, when the RSSI value of the radio wave signal from the actual target article T whose position is to be specified is measured in each receiver AP1 to APn, the measurement data is obtained. Then, the position of the target article T is estimated by comparing with the pre-learning data stored in advance in the database of the position estimation device LS. Here, the radio wave signal transmitted by the target article T has the same signal system and frequency band as the frequency band of the test radio wave signal transmitted by the test transmitters PL1 to PLm used for collecting the pre-learning data. is there.

  In general, in the pre-learning method as shown in FIG. 9, in order to improve the estimation accuracy of the position of the target article T, the test transmitters PL1 to PLm prepared in advance for collecting the pre-learning data are used. It is necessary to measure in advance the RSSI value of the test radio wave signal at the location and store more pre-learning data in the database of the position estimation device LS. However, in order to store more pre-learning data in the database, there is a problem that more resources such as the amount of measurement work and the amount of database are required.

P.Bahl, and V.N.Padmanabhan, “RADAR: An In-building RF-based User Location and Tracking System”, Proceedings of IEEE Infocom2000, vol.2, pp.775-784.

  The present invention has been made in view of the circumstances as described above, and when applying a pre-learning method, the number of test transmitters prepared in advance for pre-learning and the test transmitters arranged for pre-learning Even when the number of indoor placement locations and the number of receivers placed in specific indoor locations are small, the location of the target article placed indoors can be identified more accurately with less computation An object of the present invention is to provide an article position specifying system, an article position specifying method, and an article position specifying program.

  The present invention comprises the following technical means in order to solve the above-mentioned problems.

The first technical means is an article position specifying system for specifying the position of an article installed indoors,
An RSSI value (Received Signal Strength Indicator) regarding a test radio wave signal transmitted from each of one or a plurality of test transmitters arranged in one or a plurality of indoor locations arbitrarily set for pre-learning ) Is continuously collected for a predetermined period of time by each of one or more receivers installed in each of one or more predetermined specific locations;
The indoor location where the test transmitter is not arranged by performing a regression process by a Gaussian process on the RSSI value of the test radio wave signal collected for pre-learning by the pre-learning data collection unit A pre-learning RSSI value map generating unit that estimates the RSSI value of the test radio wave signal when transmitted from the mobile station and generates a pre-learning RSSI value map that covers the entire indoor space;
The RSSI value of the radio wave signal transmitted from the target article whose position is to be identified among the articles installed indoors is respectively determined by one or more of the receivers installed at one or more of the specific locations. A target article RSSI value collection unit for collecting the target article RSSI value;
Target for estimating the position of the target article based on the target article RSSI value collected by the target article RSSI value collection unit and the pre-learning RSSI value map generated by the pre-learning RSSI value map generation unit An article position estimation unit;
It is characterized by having at least.

The second technical means is the article position specifying system according to the first technical means,
The pre-learning RSSI value map generator generates the pre-learning RSSI value map, the test radio wave signal continuously collected for the predetermined time by the pre-learning data collection unit. The pre-learning RSSI value covering the entire spatial area of the indoor space by performing a regression process by a Gaussian process on the discrete probability density function calculated with respect to the RSSI value and converting it into a continuous probability density function. The map is generated as a probability distribution map of continuous RSSI values.

The third technical means is the article position specifying system according to the first or second technical means,
When the target article position estimation unit identifies the position of the target article based on the target article RSSI value and the pre-learning RSSI value map, the target article RSSI value and the pre-learning RSSI value map On the other hand, a maximum likelihood method is applied to estimate a position most similar to the target article RSSI value.

The fourth technical means is an article position specifying method for specifying the position of an article installed indoors,
An RSSI value (Received Signal Strength Indicator) regarding a test radio wave signal transmitted from each of one or a plurality of test transmitters arranged in one or a plurality of indoor locations arbitrarily set for pre-learning ) Is continuously collected for a predetermined period of time by each of one or more receivers installed in each of one or more predetermined specific locations;
The indoor location where the test transmitter is not arranged by performing a regression process by a Gaussian process on the RSSI value of the test radio wave signal collected for pre-learning in the pre-learning data collection step A pre-learning RSSI value map generating step of estimating an RSSI value of the test radio wave signal when transmitted from the base station and generating a pre-learning RSSI value map that covers the entire indoor spatial area;
The RSSI value of the radio wave signal transmitted from the target article whose position is to be identified among the articles installed indoors is respectively determined by one or more of the receivers installed at one or more of the specific locations. A target article RSSI value collecting step of collecting as a target article RSSI value;
Target for estimating the position of the target article based on the target article RSSI value collected in the target article RSSI value collection step and the pre-learning RSSI value map generated in the pre-learning RSSI value map generation step An article position estimating step;
It is characterized by having at least.

The fifth technical means is the article position specifying method according to the fourth technical means,
In the pre-learning RSSI value map generation step, when the pre-learning RSSI value map is generated, the test radio wave signals collected continuously during the predetermined time in the pre-learning data collection step The pre-learning RSSI value covering the entire spatial area of the indoor space by performing a regression process by a Gaussian process on the discrete probability density function calculated with respect to the RSSI value and converting it into a continuous probability density function. The map is generated as a probability distribution map of continuous RSSI values.

  A sixth technical means is a Squared as a covariance function applied when the discrete probability density function is converted into the continuous probability density function in the article position specifying method according to the fifth technical means. One of Exponential covariance function, Matern covariance function, and Rational Quadratic covariance function is applied.

A seventh technical means is the article position specifying method according to any one of the fourth to sixth technical means,
In the target article position estimation step, when the position of the target article is specified based on the target article RSSI value and the pre-learning RSSI value map, the target article RSSI value and the pre-learning RSSI value map are used. On the other hand, a maximum likelihood method is applied to estimate a position most similar to the target article RSSI value.

  The eighth technical means is an article position specifying program that implements the article position specifying method according to any one of the fourth to seventh technical means as a program executable by a computer. .

  According to the article position specifying system, the article position specifying method, and the article position specifying program of the present invention, the following effects can be obtained.

  That is, the RSSI value of the radio wave signal transmitted from the test transmitter for pre-learning placed in each of one or more known locations arbitrarily determined for pre-learning is collected as pre-learning data, and the collected By applying regression analysis by Gaussian process to the pre-learning data and preparing the pre-learning RSSI value map in advance as a probability distribution map of continuous RSSI values, The position of the target article can be estimated and specified more accurately even if the amount of pre-learning data is significantly smaller than that of the prior learning technique.

It is a system configuration figure showing an example of the system configuration of the article position specification system concerning the present invention. It is a flowchart for demonstrating an example of operation | movement of the article | item position identification system of FIG. It is explanatory drawing for demonstrating an example of the specific process sequence which produces | generates the pre-learning RSSI value map by the data measurement part for pre-learning of FIG. It is explanatory drawing explaining the Squared Exponential covariance function applied as a covariance function to the Gaussian process regression process which produces | generates a prior learning RSSI value map. It is explanatory drawing explaining the Matern covariance function applied as a covariance function to the Gaussian process regression process which produces | generates a prior learning RSSI value map. It is explanatory drawing explaining the Rational Quadratic covariance function applied as a covariance function to the Gaussian process regression process which produces | generates a prior learning RSSI value map. It is explanatory drawing for demonstrating an example of the specific process sequence of the maximum likelihood method applied as a position specific algorithm for estimating the position of the target article T by the target article position estimation part of FIG. It is a schematic diagram for demonstrating the method of specifying the position of an article | item from the RSSI value of the conventional radio wave signal. It is a schematic diagram for demonstrating the estimation method of the RSSI value of the radio wave signal by the conventional prior learning system.

  Hereinafter, an example of a preferred embodiment of an article position specifying system, an article position specifying method, and an article position specifying program according to the present invention will be described in detail with reference to the drawings. In the following description, the article position specifying system and the article position specifying method according to the present invention will be described. However, the article position specifying method may be implemented as an article position specifying program executable by a computer. Furthermore, it goes without saying that the article position specifying program may be recorded on a computer-readable recording medium.

(Features of the present invention)
Prior to the description of the embodiments of the present invention, an outline of the features of the present invention will be described first. The present invention relates to a technology for specifying the position of a target article that emits a radio wave signal by mounting an active radio tag, a radio device, or the like, and uses the radio wave signal emitted from the target article. Pre-learning data obtained by pre-learning in order to specify the position of the target article (that is, RSSI relating to a test radio wave signal having the same signal system, frequency band and electric field intensity as the radio wave signal emitted by the target article) By applying a Gaussian regression analysis method to the value (Received Signal Strength Indicator), a probability distribution map for continuous RSSI values is created in advance as a pre-learning RSSI value map. The main feature is that it is possible to estimate and identify the exact position of the target article with less calculation amount. There is a special feature.

  Conventionally, as described in Non-Patent Document 1, various methods have been proposed for identifying and tracking the position information of an article using a wireless tag or the like.

  In this, the RSSI value of the radio wave signal that estimates the position of the target (target article) placed indoors using the RSSI value (Received Signal Strength Indicator) of the radio wave signal was used. The article location method has an advantage that an accurate RSSI value can be obtained even when a simple measuring device (sensor) is used as a receiver. On the other hand, the radio wave signal is affected by noise, multipath, etc. This is accompanied by a demerit that it is difficult to accurately estimate the position of the target article.

  Therefore, in order to accurately identify the position of the target article based on the RSSI value relating to the received radio wave signal, as described in Non-Patent Document 1, a test transmitter prepared for prior learning is used in advance. A prior learning method is proposed in which an RSSI map in which the correspondence between the RSSI value of the radio wave signal transmitted from the test transmitter and the position of the test transmitter is registered in advance is created by performing the prior learning. Has been.

  However, the conventional pre-learning method like this has a problem in carrying out the pre-learning as described above. In other words, in order to improve the estimation accuracy of the position of the target article, it is necessary to measure the RSSI values of the radio wave signals at a larger number of points in advance and to store more pre-learning data in the database. Therefore, the work efficiency in the pre-learning phase is poor, and there is a problem that more resources such as the work amount of the measurement and the database amount are required.

  On the other hand, in the present invention, by applying a regression analysis method using a Gaussian process to the pre-learning data, a probability distribution map relating to RSSI values that continuously cover the entire indoor area is converted into the pre-learning RSSI. It is possible to generate a value map in advance, and thus the position of the target article whose position is to be identified with a smaller amount of calculation even when the pre-collected pre-learning data is less. Can be identified with a more accurate estimation.

(Embodiment of the present invention)
Next, an example of the embodiment of the article position specifying system according to the present invention will be described. FIG. 1 is a system configuration diagram showing an example of a system configuration of an article position specifying system according to the present invention.

  As shown in FIG. 1, the article position specifying system includes a position estimation device LS provided with a database capable of storing various data, and a radio from a transmitter (such as a wireless tag or a wireless device) provided for an indoor article. One or more receivers (access points) AP1 to APn (n = 5 in the case of FIG. 1) installed in one or more predetermined specific locations indoors as sensors for receiving radio signals, for pre-learning 1 to 1 or a plurality of test transmitters (test access points) PL1 to PLm (in the case of FIG. 1), which are arranged at one or more known indoor locations arbitrarily set as , M = 9), and at least a target article T to be a target for specifying a position among articles installed indoors. The target article T is attached with a transmitter (wireless tag, wireless device, etc.) that transmits a radio wave signal for specifying a position.

  Here, the receivers (access points) AP1 to APn are respectively arranged at one or more predetermined specific locations indoors where the target article T for estimating the position is arranged, and are used for pre-learning. A test radio wave signal transmitted from each of the one or more test transmitters PL1 to PLm and a radio radio signal transmitted from the target article T after prior learning are received. Here, the locations of the receivers (access points) AP1 to APn are specific locations that are predetermined as known locations as described above.

  Further, the test radio wave signals transmitted from one to a plurality of test transmitters (test access points) PL1 to PLm used for pre-learning are equivalent to the radio radio signals transmitted from the target article T after the pre-learning. This signal has a signal system, frequency band, and electric field strength. Further, m test transmitters (access points for testing) PL1 to PLm (m = 9 in the case of FIG. 1) are each in an indoor area where the target article T whose position is to be specified may be placed. Are arranged at one or more known locations (m = 9 in the case of FIG. 1) that are arbitrarily selected uniformly and transmit test radio wave signals. The transmitter is an easily movable transmitter, and the test radio wave signal may be transmitted while sequentially moving to an arbitrarily selected location without arranging m units in advance. Note that the test transmitters (test access points) PL1 to PLm are removed from the indoors when the collection of the pre-learning data for pre-learning is completed.

  Further, as described above, the target article T as a target is equipped with a function of transmitting a radio wave signal such as an active wireless tag or a wireless device, and when the target article T is placed indoors, Therefore, it is possible to continuously transmit a radio wave signal.

  In addition, as shown in FIG. 1, the position estimation device LS provided with a database is a pre-learning data collection unit 11, a pre-learning RSSI value, as an arithmetic processing unit that performs various types of information processing for estimating the position of an article. At least a map generation unit 12, a target article RSSI value collection unit 13, and a target article position estimation unit 14, and as a database for storing various information, a pre-learning data storage table 21, a pre-learning RSSI value map At least a registration table 22 and a target article RSSI value storage table 23 are provided.

  The pre-learning data collection unit 11 arranges the temporal change state of the RSSI value of the test radio wave signal transmitted from each of one or more test transmitters PL1 to PLm for pre-learning at a specific location. Are continuously received for a predetermined time (for example, 5 minutes) by each of the one or more receivers AP1 to APn, and stored in the pre-learning data storage table 21 of the database as pre-learning data. Perform accumulation processing.

  The pre-learning RSSI value map generator 12 collects the pre-learning data storage table 21 for each of one or a plurality of receivers AP1 to APn and stores them in the pre-learning data storage table 21 for pre-learning. Statistical characteristics of the RSSI value of the test radio wave signal stored in each of the receivers AP1 to APn accumulated as data are extracted as a probability density distribution and subjected to regression processing by a Gaussian process.

  Thus, based on the RSSI values of the test radio wave signals from known locations where the test transmitters (test access points) PL1 to PLm are respectively arranged, the test transmitters (test access points) PL1 to PLm By estimating the RSSI value of the test radio wave signal when the signal is transmitted from an indoor spatial location where no signal is placed, not only the RSSI value at a spatially discrete location indoors but also the indoor space An RSSI value at a continuous location that covers the entire region with a probability distribution is generated and set and registered in the pre-learning RSSI value map registration table 22 of the database as a pre-learning RSSI value map.

  Therefore, as a result of prior learning, even if the amount of collected RSSI values of the test radio wave signals using the test transmitters (test access points) PL1 to PLm is small, the target article T whose position is to be specified. In order to estimate the position of the target article T more accurately with a smaller amount of calculation, it is possible to generate in advance an RSSI value that continuously covers the entire indoor area necessary for estimating the position relating to the position as a probability distribution. Will be ready.

  Further, the target article RSSI value collection unit 13 sets one or more RSSI values of radio wave signals transmitted from the target article T whose position is to be identified among articles installed indoors after prior learning. Are received by each of one or more receivers AP1 to APn arranged at each of the specific locations, and stored in the target article RSSI value storage table 23 of the database as the target article RSSI value.

  In addition, the target article position estimation unit 14 collects the RSSI value of the radio wave signal from the target article T collected by the target article RSSI value collection unit 13 and stored in the target article RSSI value storage table 23 of the database, that is, the target article RSSI value. And data processing such as maximum likelihood (MLE) based on the pre-learning RSSI value map generated by the pre-learning RSSI value map generation unit 12 and registered in the pre-learning RSSI value map registration table 22 Is applied to estimate the position of the target article T.

  Note that the position estimation device LS may be constructed as a position-specific control device equipped with a database, or may be constructed as a general-purpose computer equipped with a database. In addition, each part of the pre-learning data collection unit 11, the pre-learning RSSI value map generation unit 12, the target article RSSI value collection unit 13, and the target article position estimation unit 14 that constitutes the arithmetic processing unit of the position estimation device LS is provided. The pre-learning data collection unit 11, the pre-learning RSSI value map generation unit 12, may not be arranged in the same control unit or computer, and may be distributed to a plurality of different control units or computers. One or more functions (including all functions) among the functions performed by each part of the target article RSSI value collection unit 13 and the target article position estimation unit 14 are controlled by a control computer, a general-purpose computer, or the like. It can also be configured as a program executable by a computer.

  Further, the database provided in the position estimation device LS is also configured as a database server independent of the position estimation device LS, and the pre-learning data collection unit 11 and the target article RSSI value collection unit 13 are included in the database server. The same function may be arranged together, but in any configuration, the database is transmitted by the test collection collected by the pre-learning data collection unit 11, the target article RSSI value collection unit 13, etc. It has the capability of accumulating data related to the RSSI value of the test radio wave signal from the machines PL1 to PLm and the RSSI value of the radio wave signal from the target article T in a predetermined area in real time.

(Operation example of the embodiment of the present invention)
Next, an example of the operation of the article position specifying system according to the present invention shown in FIG. 1 will be described with reference to the flowchart of FIG. FIG. 2 is a flowchart for explaining an example of the operation of the article position specifying system in FIG. 1, and shows an example of the article position specifying method according to the present invention.

  As shown in the flowchart of FIG. 2, the article position specifying method according to the present invention uses the RSSI values of test radio wave signals transmitted from any one or more predetermined indoor positions for pre-learning. A pre-learning phase (flowchart in FIG. 2A) in which data is collected and pre-learned, and an RSSI value of a radio wave signal transmitted from the target article T to be actually identified after pre-learning and pre-learning Based on the learning result, it is divided into two phases, a position estimation phase for estimating the position of the target article T (the flowchart in FIG. 2B), and the following configuration requirements It is made up of.

  First, the pre-learning phase of FIG. 2 (A) includes the following two steps: Step 1 (pre-learning data collection step) and Step 2 (pre-learning RSSI value map generation step).

(1) Collection of pre-learning data by the pre-learning data collection unit 11 (step Step 1 in FIG. 2A)
The pre-learning data collection unit 11 shown in FIG. 1 includes one or more receivers AP1 to APn (in FIG. 1) installed at one or more predetermined specific locations indoors as sensors for receiving radio wave signals. In this case, n = 5), one or more test transmitters arranged at one or more known locations arbitrarily set in the room for pre-learning for a predetermined time (for example, 5 minutes). (Access point for test) PL 1 to PLm (in the case of FIG. 9, m = 9) Continuously measure the RSSI value (received radio wave intensity) of the test radio wave signal transmitted from each, and the data for pre-learning Is stored in the pre-learning data storage table 21 of the database.

(2) Generation of a pre-learning RSSI value map based on the analysis result of the pre-learning data by the pre-learning RSSI value map generation unit 12 (step Step 2 in FIG. 2A)
Next, the pre-learning RSSI value map generation unit 12 in FIG. 1 performs regression analysis using a Gaussian process on the pre-learning data stored in the pre-learning data storage table 21 by the pre-learning data collection unit 11 in Step 1. By applying the method, the RSSI value (received electric field strength) of the radio wave signal from other indoor locations (that is, the location where the test transmitter for transmitting the test radio wave signal is not arranged) is continuously obtained. Estimated as a probability distribution, a pre-learning RSSI value map that covers the entire indoor space continuously from both the place where the test transmitter is placed and the place where the test transmitter is not placed An RSSI value probability distribution map is generated and set and registered in the pre-learning RSSI value map registration table 22 of the database.

  In the position estimation phase that is activated after the pre-learning RSSI value map that continuously covers the entire indoor area is set and registered in advance in the pre-learning RSSI value map registration table 22 of the database by the pre-learning phase like this, The following Step 3 (target article RSSI value collection step) and Step 4 (target article position estimation step) are included.

(3) Collection of RSSI values of radio wave signals from the target article T by the target article RSSI value collection unit 13 (Step 3 in FIG. 2B)
When the target article RSSI value collection unit 13 of FIG. 1 wants to specify the position of a target placed indoors, that is, the target article T, by actually transmitting a radio wave signal from the target article T placed indoors, The RSSI value of the radio wave signal is collected by each of the receivers AP1 to APn arranged as sensors at one or more predetermined specific locations, and stored in the target article RSSI value storage table 23 of the database.

(4) Estimation of the position of the target article by the target article position estimation unit 14 (Step Step 4 in FIG. 2B)
The target article position estimation unit 14 in FIG. 1 performs the RSSI value of the radio wave signal from the target article T stored in the target article RSSI value storage table 23 by the target article RSSI value collection unit 13 in step Step 3 and step 2 in the pre-learning phase. Based on the pre-learning RSSI value map set and registered in the pre-learning RSSI value map registration table 22 by the pre-learning RSSI value map generation unit 12, data processing based on a location specifying algorithm such as maximum likelihood (MLE) is applied. Thus, the position of the target article T whose position is to be specified as a target is estimated.

  Next, an example of a specific processing procedure for generating the pre-learning RSSI value map by the pre-learning RSSI value map generation unit 12 in Step 2 of FIG. 2 will be further described with reference to FIG. FIG. 3 is an explanatory diagram for explaining an example of a specific processing procedure for generating a pre-learning RSSI value map by the pre-learning RSSI value map generation unit 12 of FIG. 1.

  As shown in FIG. 3, the RSSI values of the test radio wave signals from the test transmitters PL1 to PLm that are arranged at m locations that are arbitrarily determined in advance for pre-learning are determined in specific locations. Is continuously measured at a predetermined period for a predetermined time (for example, 5 minutes) by the receivers AP1 to APn installed in the receiver, and the pre-learning data RSS_Cal1 for each of the receivers AP1 to APn. , RSS_Cal2,..., RSS_CalN.

After that, for each of the collected pre-learning data RSS_Cal1, RSS_Cal2,. As statistical characteristics of the pre-learning data of each of the transmitters PL1 to PLm, that is, RSSI value data, a probability density function pdf (Probability Density Function), that is, pdf (RSS / x _i ) (1)
Where i = 1, 2,..., N
Are extracted as discrete RSSI value data for each of the n receivers AP1 to APn.

Here, the probability density function pdf (RSS / x _i ) indicating the statistical characteristics of the RSSI value data of the test transmitters PL1 to PLm extracted for each of the receivers AP1 to APn is spatially discrete. It is RSSI value data. Therefore, for each receiver AP1 to APn, against discrete probability density function pdf (RSS / _{x i)} about RSSI value data for each test transmitter PL1～PLm, by performing a regression process by Gaussian process Then, continuous RSSI value data based on a continuous probability distribution is obtained.

That is, as shown in the following equation (2), a discrete probability density function pdf (RSS / x _i ) is covered as a probability distribution over the entire indoor area by Gaussian process regression processing for each of the receivers AP1 to APn. To a continuous probability density function pdf (RSS / x) indicating a continuous RSSI value.

pdf (RSS / x) = n (μ _RSS (x), σ ² _RSS (x)) (2)
Here, μ: average value of RSSI value, σ ² : variance of RSSI value The probability distribution map of pdf (RSS / x) generated by Gaussian process regression processing is a pre-learning RSSI value map registration table as a pre-learning RSSI value map. 22 is set and registered.

  In addition, in the lower left of FIG. 5, a distribution state of discrete RSSI values in each indoor location where the test transmitters PL <b> 1 to PLm are arranged is schematically shown, and in the lower right of FIG. 5, A cover image of continuous RSSI values generated as a continuous probability distribution by Gaussian process regression processing is schematically shown. However, in the lower right cover image of FIG. 5, the RSSI value changes in a stepped manner and is expressed as if there is a boundary line between different RSSI values, but the RSSI value obtained by Gaussian process regression processing is It is a continuous and gradual change, and no such boundary exists.

Here, as a covariance function applied when the Gaussian process regression process is performed on the probability density function pdf (RSS / x _i ) indicating the statistical characteristics of the RSSI value data of the test transmitters PL1 to PLm, Any of Squared Exponential (SE: square exponential covariance function), Matern (general nonstationary covariance function), Rational Quadratic (RQ: rational quadratic covariance function), etc. as shown in FIGS. A covariance function may be applied.

  FIG. 4 is an explanatory diagram for explaining a Squared Exponential covariance function applied as a covariance function to a Gaussian process regression process for generating a pre-learning RSSI value map, and FIG. 5 is a Gauss process for generating a pre-learning RSSI value map. FIG. 6 is an explanatory diagram for explaining a Matern covariance function applied as a covariance function to a regression process. FIG. 6 shows a Rational Quadratic covariance function applied as a covariance function to a Gaussian process regression process for generating a pre-learning RSSI value map. It is explanatory drawing demonstrated.

As shown in FIG. 4, the Squared Exponential covariance function k _SE (x _i , x _j ) is a smooth function, can be differentiated infinitely, and is given by the following equation (3).

k _SE (x _i , x _j ) = σ ² _f exp {− (x _i −x _j ) ² / 2l ² } + σ ² _v δ _ij
... (3)
Where σ ² _f : amplitude variance
l: Length Scale characterizing dispersion
σ ² _ν : Noise variance In equation (3), amplitude variance σ ² _f , Length Scale 1 and noise variance σ ² _ν can be set to arbitrary values manually, but marginal likelihood p ( It is also possible to derive from the one that maximizes y / x, l, σ ² _f , σ ² _v ).

Further, as shown in FIG. 5, the Matern covariance function k _Matern (x _i , x _j ) is given by the following equation (4).

k _Matern (x _i , x _j ) = (2 ^1−ν / Γ (ν)) · ((2ν) ^1/2 | x _i −x _j | / l) · Κ _ν ((2ν) ^1/2 | x _i −x _j | / l) (4)
Where ν is a parameter that defines smoothness (| 1-ν | is differentiable)
l: Length Scale characterizing dispersion
Γ (ν): Gamma function
_{Ν ν} : Bessel function In equation (4), the larger ν is, the smoother it is, but when ν = 1/2, it becomes Brownian motion, and when ν → ∞, the same Squared Exponential covariance as equation (1) Become a function.

Also, as shown in FIG. 6, the Rational Quadratic covariance function k _RQ (x _i , x _j ) is given by the following equation (5).

k _RQ (x _i , x _j ) = (1+ | xi−xj | ² / 2αl} ^−α (5)
Where α is a parameter that defines smoothness
l: Length Scale characterizing dispersion
In Equation (5), when ν → ∞, the Squared Exponential covariance function is the same as in Equation (1).

  Next, FIG. 7 is used as an example of a maximum likelihood method (MLE: Maximum Likelihood Estimation) applied as a position specifying algorithm for estimating the position of the target article T by the target article position estimating unit 14 in Step 4 of FIG. Further explanation will be given. FIG. 7 is an explanatory diagram for explaining an example of a specific processing procedure of the maximum likelihood method applied as a position specifying algorithm for estimating the position of the target article T by the target article position estimating unit 14 of FIG. .

When estimating the position of the target article T in the target article position estimation unit 14, as described above, the pre-learning RSSI value map generated as a pre-learning result and the RSSI value of the radio wave signal from the target article T are used. The position most similar to the RSSI value of the radio wave signal from the target article T is estimated by applying the maximum likelihood method to the target article RSSI value. That is, as shown in FIG. 7, for example, statistically independent RSSI values s = {s ₁ , respectively, from n receivers (access points) AP1 to APn that have received a radio wave signal from the target article T. , S ₂ ,..., S _n }, as an estimation method for the statistical model with unknown parameters, the estimation of the position x _MLE of the target article T is expressed by the following equation (6): It can be estimated by maximizing the joint likelihood p (x / s) for an unknown location.

ａｒｇｍａｘ：結合尤度ｐ（ｘ／ｓ）の関数を最大化するｘが存在するとき
の値
ここで、ベイズ法を適用する場合は、次の式（７）に示す関係が得られる。

argmax: when there exists x that maximizes the function of the joint likelihood p (x / s)
Here, when applying the Bayesian method, the relationship shown in the following equation (7) is obtained.

したがって、ＲＳＳＩ値の平均をＧＰ＿Ｍｅａｎ、ＲＳＳＩ値の分散をＧＰ＿Ｖａｒと表すと、次の式（８）に示すように、

Therefore, when the average of the RSSI values is expressed as GP_Mean and the variance of the RSSI values is expressed as GP_Var, as shown in the following equation (8),

の関係が成立する。

The relationship is established.

  Furthermore, the Gaussian negative log likelihood function for equation (8) is given by equation (9) below.

したがって、結合尤度が最大化する値を求めるには、この式（９）を微分した結果が０になる平均ＧＰ＿Ｍｅａｎの値を求めることによって、対象物品Ｔの位置ｘ_ＭＬＥを推定することができる。

Therefore, in order to obtain a value at which the joint likelihood is maximized, the position x _MLE of the target article T can be estimated by obtaining the average GP_Mean value at which the result obtained by differentiating the equation (9) becomes zero. .

(Effect of this embodiment)
As described above in detail, in the present embodiment, radio waves transmitted from the test transmitters PL1 to PLm for pre-learning arranged in one or more known locations arbitrarily determined in advance for pre-learning. The RSSI value of the signal is collected as pre-learning data, and regression analysis by a Gaussian process is applied to the collected pre-learning data, so that the pre-learning RSSI value map is previously obtained as a probability distribution map of continuous RSSI values. By creating the data, it is possible to obtain a special effect that the position of the target article T can be estimated more accurately even with a small amount of pre-learning data.

DESCRIPTION OF SYMBOLS 11 ... Pre-learning data collection part, 12 ... Pre-learning RSSI value map generation part, 13 ... Target article RSSI value collection part, 14 ... Target article position estimation part, 21 ... Pre-learning data accumulation table, 22 ... Pre-learning RSSI Value map registration table, 23 ... target article RSSI value storage table, AP1 to APn ... receiver (access point, sensor), LS ... position estimation device, PL1 to PLm ... test transmitter (test access point), T ... target Article (target).

Claims (8)

An article position specifying system for specifying the position of an article installed indoors,
An RSSI value (Received Signal Strength Indicator) regarding a test radio wave signal transmitted from each of one or a plurality of test transmitters arranged in one or a plurality of indoor locations arbitrarily set for pre-learning ) Is continuously collected for a predetermined period of time by each of one or more receivers installed in each of one or more predetermined specific locations;
The indoor location where the test transmitter is not arranged by performing a regression process by a Gaussian process on the RSSI value of the test radio wave signal collected for pre-learning by the pre-learning data collection unit A pre-learning RSSI value map generating unit that estimates the RSSI value of the test radio wave signal when transmitted from the mobile station and generates a pre-learning RSSI value map that covers the entire indoor space;
The RSSI value of the radio wave signal transmitted from the target article whose position is to be identified among the articles installed indoors is respectively determined by one or more of the receivers installed at one or more of the specific locations. A target article RSSI value collection unit for collecting the target article RSSI value;
Target for estimating the position of the target article based on the target article RSSI value collected by the target article RSSI value collection unit and the pre-learning RSSI value map generated by the pre-learning RSSI value map generation unit An article position estimation unit;
An article position specifying system comprising: In the article location system according to claim 1,
The pre-learning RSSI value map generator generates the pre-learning RSSI value map, the test radio wave signal continuously collected for the predetermined time by the pre-learning data collection unit. The pre-learning RSSI value covering the entire spatial area of the indoor space by performing a regression process by a Gaussian process on the discrete probability density function calculated with respect to the RSSI value and converting it into a continuous probability density function. An article location system characterized by generating a map as a probability distribution map of continuous RSSI values. In the article location system according to claim 1 or 2,
When the target article position estimation unit identifies the position of the target article based on the target article RSSI value and the pre-learning RSSI value map, the target article RSSI value and the pre-learning RSSI value map An article position specifying system that applies a maximum likelihood method to estimate a position most similar to the target article RSSI value. An article position specifying method for specifying the position of an article installed indoors,
An RSSI value (Received Signal Strength Indicator) regarding a test radio wave signal transmitted from each of one or a plurality of test transmitters arranged in one or a plurality of indoor locations arbitrarily set for pre-learning ) Is continuously collected for a predetermined period of time by each of one or more receivers installed in each of one or more predetermined specific locations;
The indoor location where the test transmitter is not arranged by performing a regression process by a Gaussian process on the RSSI value of the test radio wave signal collected for pre-learning in the pre-learning data collection step A pre-learning RSSI value map generating step of estimating an RSSI value of the test radio wave signal when transmitted from the base station and generating a pre-learning RSSI value map that covers the entire indoor spatial area;
The RSSI value of the radio wave signal transmitted from the target article whose position is to be identified among the articles installed indoors is respectively determined by one or more of the receivers installed at one or more of the specific locations. A target article RSSI value collecting step of collecting as a target article RSSI value;
Target for estimating the position of the target article based on the target article RSSI value collected in the target article RSSI value collection step and the pre-learning RSSI value map generated in the pre-learning RSSI value map generation step An article position estimating step;
An article position specifying method characterized by comprising: In the article position specifying method according to claim 4,
In the pre-learning RSSI value map generation step, when the pre-learning RSSI value map is generated, the test radio wave signals collected continuously during the predetermined time in the pre-learning data collection step The pre-learning RSSI value covering the entire spatial area of the indoor space by performing a regression process by a Gaussian process on the discrete probability density function calculated with respect to the RSSI value and converting it into a continuous probability density function. An article position specifying method, wherein the map is generated as a probability distribution map of continuous RSSI values.   6. The article location method according to claim 5, wherein as a covariance function applied when converting the discrete probability density function to the continuous probability density function, a Squared Exponential covariance function, a Matern covariance function, An article location method characterized by applying one of the Rational Quadratic covariance functions. In the article position specifying method according to any one of claims 4 to 6,
In the target article position estimation step, when the position of the target article is specified based on the target article RSSI value and the pre-learning RSSI value map, the target article RSSI value and the pre-learning RSSI value map are used. An article position specifying method characterized by applying a maximum likelihood method to estimate a position most similar to the target article RSSI value.   8. An article position specifying program, wherein the article position specifying method according to claim 4 is implemented as a program executable by a computer.

Images