JP2019174164A - Device, program and method for estimating terminal position using model pertaining to object recognition information and received electromagnetic wave information - Google Patents

Abstract

To provide a device with which it is possible to more accurately estimate the position of a terminal having received an electromagnetic wave, even in a state where a movable object exists that affects an electromagnetic wave form an electromagnetic wave source.SOLUTION: The present device for acquiring received electromagnetic wave information at a terminal having received an electromagnetic wave from an electromagnetic wave source, comprises: learning data generation means, capable of acquiring object recognition information from object recognition means capable of recognizing an object that can move, for generating feature data including received electromagnetic wave information and object recognition information at a given point of time and learning data including terminal position information pertaining to the position of a terminal at the point of time; model generation means for generating a model for terminal position estimation using a plurality of learning data; and terminal position determination means for determining, using the generated model, terminal position information pertaining to the terminal the position of which is being estimated, on the basis of feature data including received electromagnetic wave information at a position estimated point of time acquired from the terminal the position of which is being estimated and the object recognition information at a position estimated point of time.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a technique for estimating the position of a terminal based on the reception status of electromagnetic waves at the terminal.

  Conventionally, a technique for estimating the position of a receiving terminal using a plurality of received radio signals (electromagnetic waves) has become widespread. For example, positioning technology using outdoor GPS (Global Positioning System), indoor wireless LAN (Local Area Network), and the like are widely used at present.

  In such positioning technology using electromagnetic waves, whether or not there is an object that affects the propagation of electromagnetic waves between an electromagnetic wave transmission source (electromagnetic wave source) such as a satellite, a base station or an access point and the terminal, Positioning results are greatly affected. For example, when a shielding object that shields electromagnetic waves is interposed, the received electromagnetic wave intensity at the terminal is lower than when the electromagnetic waves reach the terminal directly, and the distance between the electromagnetic wave source and the terminal is estimated to be larger than actual, The error of position estimation will increase.

  As a technique for solving such a problem, for example, Non-Patent Document 1 discloses an indoor position estimation technique using both fingerprint (Fingerprint) and dead reckoning (Dead Reckoning, autonomous navigation, relative self-position estimation). Is disclosed.

  In this technique, the reception intensity of electromagnetic waves from each electromagnetic wave transmission source is learned in advance, and a database relating to the relationship between the electromagnetic wave reception intensity and the position is created in advance. Next, in the case of positioning using this database, when a plurality of terminal position candidates are estimated due to the deterioration of the received electromagnetic wave intensity due to the presence of the shielding object, the position one hour before using dead reckoning A terminal position candidate close to the coordinates is selected and adopted as the terminal position.

Kouhei Ueda, Satoshi Suzuki, Hiroshi Shigeno, “Examination of Indoor Location Estimation Method Using Fingerprint and Dead Reckoning”, IPSJ Research Reports Multimedia Communication and Distributed Processing (DPS), Vol. 2014-DPS-160 (8 ), 1-8, 2014

  In the conventional technology such as the technology of Non-Patent Document 1 described above, an object that affects the propagation of electromagnetic waves is considered, but the object to be considered is a static shield with a fixed position.・ It is an obstacle. However, in actual terminal location estimation, it is often the case that a person cannot be ignored as an object that obstructs electromagnetic wave (wireless signal) propagation.

  For example, nowadays, in stores in various fields, there is an increasing need for accurately grasping the flow lines of store staff and customers who have terminals. In this case, it is common to estimate the terminal position by arranging a plurality of beacon transmission sources (hereinafter abbreviated as beacons) in the store. Here, as a matter of course, the person moving in the store greatly affects the terminal position estimation by these beacons.

  However, in the technique described in Non-Patent Document 1, for example, a database created in advance is only applied to an environment where static obstacles installed in advance exist. It is not applicable to an environment where such a dynamic shield exists. Thus, although the influence on electromagnetic wave propagation by a movable object including a person is great, it has not been considered in the past.

  Therefore, the present invention provides an apparatus, a program, and a method that can more accurately estimate the position of a terminal that has received an electromagnetic wave even in a situation where a movable object that affects the electromagnetic wave from the electromagnetic wave source may exist. The purpose is to provide.

According to the present invention, there is provided a terminal position estimation device that acquires received electromagnetic wave information in a terminal that has received an electromagnetic wave from an electromagnetic wave source and estimates the position of the terminal,
Object recognition information from an object recognition means capable of recognizing a movable object can be acquired, the received electromagnetic wave information at a certain time of the terminal, and feature data including the object recognition information at the time, Learning data generating means for generating learning data including terminal position information relating to the position of the terminal at the time,
Model generation means for generating a model for terminal position estimation using a plurality of the learning data;
Based on the feature data including the received electromagnetic wave information at the position estimation time acquired from the position estimation target terminal and the object recognition information at the position estimation time using the model, the position estimation target terminal There is provided a terminal location estimation apparatus having terminal location determination means for determining such terminal location information.

As one embodiment of the terminal position estimation apparatus according to the present invention, the object identification information is object position information related to the position of the recognized object,
The terminal position determination unit compares the object position information at the time of the position estimation with the terminal position information that is a determination candidate derived from the model, and determines the position estimation based on the proximity of each other's position. It is also preferable to determine terminal location information related to the target terminal.

  Further, in this embodiment, the terminal position determination means includes the position related to the object position information and the terminal position information that is a determination candidate derived from the model among the object position information at the time of the position estimation. It is also preferable to determine the object position information having the shortest distance to the position related to the above or a predetermined threshold value or less as the terminal position information related to the position estimation target terminal.

  Furthermore, in the above-described embodiment, the terminal position determining unit includes the position related to the terminal position information and the object position at the position estimation time among the terminal position information that is a determination candidate derived from the model. It is also preferable to determine the terminal position information having the shortest distance from the position related to the information or equal to or less than a predetermined threshold as the terminal position information related to the position estimation target terminal.

  Furthermore, in the above embodiment, the terminal position determining means calculates a reliability of determination for each terminal position information that is a determination candidate derived from the model, and the terminal has the reliability equal to or higher than a predetermined threshold. It is also preferable to determine the terminal position information related to the position estimation target terminal by comparing only the position information with the object position information at the position estimation time.

Further, as another embodiment of the terminal position estimation apparatus according to the present invention,
The terminal position determining means calculates a reliability of determination for each terminal position information that is a determination candidate derived from the model,
The learning data generating means generates the learning data including the terminal position information as correct terminal position information when the determined terminal position information has the reliability equal to or higher than a predetermined threshold.
The model generation means preferably generates or updates the model using the learning data.

Furthermore, as still another embodiment of the terminal location estimation apparatus according to the present invention,
The learning data generating means generates the learning data further including the terminal position information related to the position of the terminal at the time before the time and / or the received electromagnetic wave information at the time before the time of the terminal. And
The terminal position determining means is the terminal position information related to the position at the time before the position estimation time at the position estimation target terminal and / or the time at the time before the position estimation time at the terminal of the position estimation target. It is also preferable to determine the terminal position information related to the position estimation target terminal based on the feature data further including the received electromagnetic wave information.

Furthermore, as still another embodiment of the terminal location estimation apparatus according to the present invention,
The terminal position determining means calculates a reliability of determination for each terminal position information that is a determination candidate derived from the model,
The learning data generation means generates the learning data further including the reliability calculated for the terminal position information related to the time point before the time point related to the terminal position information included in the generated learning data,
Based on the reliability calculated for the terminal position information related to the position at the time before the position estimation time at the terminal of the position estimation target, the terminal position determination unit determines whether the terminal of the position estimation target is based on the reliability. It is also preferable to determine such terminal location information.

As still another embodiment of the terminal position estimation apparatus according to the present invention, the terminal position determination means calculates a reliability of determination for each terminal position information that is a determination candidate derived from the model, and has a predetermined threshold value. When there are a plurality of the terminal location information having the reliability described above,
(A) For terminal position information determined at a time point before the position estimation time point, or (b) For terminal position information predicted from a terminal position information group determined at a past time point of the position estimation time point,
It is also preferable to select the terminal position information having the proximity of the position satisfying the predetermined condition and determine the terminal position information related to the position estimation target terminal.

  Furthermore, the model generation means of the terminal position estimation apparatus according to the present invention preferably generates a terminal position estimation model by machine learning as the model. It is also preferable to generate a terminal location database generated by a plurality of learning data.

  In the terminal position estimation apparatus according to the present invention, the object recognition information may be captured from at least one of a camera, a omnidirectional camera, a depth camera, an infrared camera, an infrared positioning unit, a laser positioning unit, and a thermographic unit. It is also preferable that it is generated based on the measurement information.

According to the present invention, there is also provided an electromagnetic wave source position estimation device for acquiring received electromagnetic wave information in a receiving device capable of receiving an electromagnetic wave from a movable electromagnetic wave source and estimating the position or position distribution of the electromagnetic wave source,
It is possible to acquire the object recognition information from the holding body that can move while holding the electromagnetic wave source, or the object recognition means that can recognize the electromagnetic wave source, the received electromagnetic wave information at a certain point of the receiving device, and Learning data generating means for generating learning data including feature data including the object recognition information at the time and electromagnetic wave source position or position distribution information related to the position or position distribution of the electromagnetic wave source at the time;
Model generation means for generating a model for electromagnetic wave source position or position distribution estimation using a plurality of the learning data,
Using the model, based on feature data including the received electromagnetic wave information at the position or position distribution estimation time and the object recognition information at the position or position distribution estimation time, the position or position distribution estimation time There is provided an electromagnetic wave source position estimating device having electromagnetic wave source position determining means for determining position or position distribution information related to the electromagnetic wave source.

According to the present invention, there is further provided a program for causing a computer mounted in an apparatus for acquiring received electromagnetic wave information in a terminal that has received an electromagnetic wave from an electromagnetic wave source and estimating the position of the terminal to function,
Object recognition information from an object recognition means capable of recognizing a movable object can be acquired, the received electromagnetic wave information at a certain time of the terminal, and feature data including the object recognition information at the time, Learning data generating means for generating learning data including terminal position information relating to the position of the terminal at the time,
Model generation means for generating a model for terminal position estimation using a plurality of the learning data;
Based on the feature data including the received electromagnetic wave information at the position estimation time acquired from the position estimation target terminal and the object recognition information at the position estimation time using the model, the position estimation target terminal There is provided a terminal position estimation program for causing a computer to function as terminal position determining means for determining such terminal position information.

According to the present invention, there is further provided a terminal position estimation method by a computer mounted on a device that acquires received electromagnetic wave information in a terminal that has received an electromagnetic wave from an electromagnetic wave source and estimates the position of the terminal,
Object recognition information from an object recognition means capable of recognizing a movable object can be acquired, the received electromagnetic wave information at a certain time of the terminal, and feature data including the object recognition information at the time, Generating learning data including terminal position information related to the position of the terminal at the time,
Generating a model for terminal position estimation using a plurality of the learning data;
Based on the feature data including the received electromagnetic wave information at the position estimation time acquired from the position estimation target terminal and the object recognition information at the position estimation time using the model, the position estimation target terminal A terminal location estimation method comprising the step of determining such terminal location information.

  According to the terminal position estimation apparatus, program, and method of the present invention, the position of the terminal that has received the electromagnetic wave can be estimated with higher accuracy even in a situation where there is a movable object that affects the electromagnetic wave from the electromagnetic wave source. It becomes possible.

It is a schematic diagram which shows one Embodiment of the terminal location estimation system containing the terminal location estimation apparatus by this invention. It is a functional block diagram which shows the function structure in one Embodiment of the terminal position estimation apparatus by this invention. It is a schematic diagram which shows roughly the terminal position estimation process by this invention including a learning phase, a position estimation phase, and a model update phase. 5 is a flowchart schematically showing an embodiment of a position estimation phase in a terminal position estimation method according to the present invention.

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

[Terminal position estimation system]
FIG. 1 is a schematic diagram showing an embodiment of a terminal location estimation system including a terminal location estimation apparatus according to the present invention.

  The terminal position estimation system of this embodiment shown in FIG. 1 estimates the momentary position in one store in at least one terminal 3 possessed by a store clerk, a customer or the like, and possesses the terminal 3 This makes it possible to grasp the flow line in the store for store clerk and customers. In the present embodiment, objects that affect electromagnetic wave propagation, such as desks, chairs, partitions, exhibition stands, and counters, are arranged in the store that is the position estimation region.

Specifically, this terminal location estimation system
(A) Beacon 2 installed at a plurality of positions in the store and transmitting electromagnetic waves at predetermined time intervals in all directions;
(B) one or a plurality of cameras 4 capable of photographing a movable object such as a person and transmitting information of the photographed image in time series via a wireless or wired communication network;
(C) “Object recognition information” is generated from a time-series image group acquired from the camera 4 via the communication network, and further directly (or via a relay device such as an access point) from the terminal 3 via the wireless communication network. ) The terminal position estimation device 1 that acquires “received electromagnetic wave information”, estimates the momentary position of the terminal 3 by machine learning based on the generated / acquired information, and determines the flow line in the store. Yes.

  Here, the beacon 2 can be, for example, a BLE (Bluetooth (registered trademark) Low Energy) specification beacon that transmits an advertisement packet in all directions at intervals of 100 to 1000 msec. Of course, the beacon 2 is not limited to this. For example, it can be used as a wireless LAN access point such as Wi-Fi (registered trademark). In addition, the terminal 3 receives electromagnetic waves (radio waves, infrared rays, etc.) therefrom and generates “received electromagnetic wave information”. If possible, various electromagnetic wave sources (radio wave sources, infrared sources, etc.) can be used as the beacon 2.

  Furthermore, in the present embodiment, the terminal 3 is a smartphone, a mobile phone, a tablet computer, a wearable terminal, or the like that can be carried by a store clerk, a customer, or the like. Applicable.

Similarly, in FIG. 1, a terminal position estimation device 1 is a device that acquires “received electromagnetic wave information” in a terminal 3 that has received an electromagnetic wave from a beacon 2 that is an electromagnetic wave source, and estimates the position of the terminal 3. As its features,
(A) It is possible to acquire “object recognition information” from an object recognition means capable of recognizing a movable object, and “received electromagnetic wave information” at a certain point in time of the terminal 3 and “object recognition information” at that point in time A learning data generation unit 114 that generates learning data including the feature data including the “terminal position information” related to the position of the terminal 3 at the time point;
(B) a model generation unit 115 that generates a “model” for terminal position estimation using a plurality of learning data;
(C) Using the generated “model”, feature data including “received electromagnetic wave information” obtained from the position estimation target terminal 3 at the time of position estimation and “object recognition information” at the time of position estimation And a terminal position determination unit 116 that determines terminal position information related to the terminal 3 that is the position estimation target.

That is, the terminal position estimation device 1
<Learning phase>: Generate a “model” for terminal position estimation from feature data including “received electromagnetic wave information” and “object recognition information” and learning data including “terminal position information” as a correct answer,
<Position estimation phase>: Using the generated “model”, the position of the terminal 3 as the position estimation target is estimated based on the feature data at the time of position estimation.

  As described above, the terminal position estimation device 1 uses the “model” generated based on the “object recognition information” including the information of the movable object, and thus the movement that affects the electromagnetic wave from the electromagnetic wave source (beacon 2). Even in a situation where a possible object may exist, the position of the terminal 3 that has received the electromagnetic wave can be estimated with higher accuracy.

  Here, in the present embodiment, objects that affect electromagnetic wave propagation such as desks, chairs, partitions, exhibition stands, and counters are arranged in the store that is the position estimation region. The terminal position estimation apparatus 1 can estimate the terminal position without acquiring the arrangement information of the objects installed in the store in advance or the arrangement information of the beacon 2.

  Therefore, even when the electromagnetic wave source (beacon 2) is mounted on, for example, the terminal 3 and can move (the “object recognition information” can identify the movable electromagnetic wave source, or “object recognition information”). The position of the terminal 3 can also be estimated, although it is assumed that all the objects identified by " Furthermore, the terminal position estimation apparatus 1 can estimate the terminal position without performing calculation processing such as calculation of the distance between the electromagnetic wave source and the terminal and calculation of the attenuation degree of the received electromagnetic wave, which can be enormous. .

  Conventionally, when calculating the distance between the electromagnetic wave source and the terminal and the attenuation of the received electromagnetic wave, since usually only the two-dimensional arrangement of the object is used, the influence of the height of the electromagnetic wave shielding object cannot be considered. On the other hand, the terminal position estimation apparatus 1 can generate a “model” including the influence of the three-dimensional arrangement of objects, and as a result, higher position estimation accuracy is realized.

  The object recognizing means having the above configuration (A) may be a functional component in the apparatus 1 as shown as an object recognizing unit 111 in FIG. 2 later, or provided in an external apparatus. (In this case, “object recognition information” is acquired from this apparatus). In any case, this object recognizing means may be able to track a moving object recognized in the image by assigning a video flow line ID (identifier).

  However, in this case as well, the terminal position estimation device 1 does not take into consideration which video flow line ID the terminal whose position is to be estimated is associated with, but also includes a “model” that incorporates “object identification information”. Can be used to determine the location of this terminal. That is, it is possible to perform the terminal position estimation process more easily without requiring a matching technique between the terminal ID and the video flow line ID.

  Further, in the terminal position estimation apparatus 1, based on the generated “model”, for example, the feature data at each time in the position estimation period can be used to estimate the position of the terminal 3 in real time almost in real time. As a result, for example, even under a situation where a large number of persons move around the store and affect the electromagnetic waves from the beacon 2, it is possible to determine the flow line of each terminal 3 with higher accuracy.

  Incidentally, the above “received electromagnetic wave information” is, for example, information in which the received signal strength (Received Signal Strength Indicator, RSSI) in the electromagnetic wave received by the terminal 3 from the beacon 2 is associated with the ID (and reception time) of the beacon 2. It can be. The “object recognition information” can be, for example, the foot position coordinates of the recognized person at each time.

  Furthermore, as another embodiment, the “object recognition information” is obtained from at least one of an omnidirectional (omnidirectional) camera, a depth camera, an infrared camera, an infrared positioning unit, a laser positioning unit, and a thermographic unit, or It may be generated based on the measurement information. For example, the depth value information (for each target pixel) generated by the depth camera may be converted into image data and used in the same manner as a captured image by the camera 4. In any case, the terminal position estimation apparatus 1 can acquire “object recognition information” generated by some means and use it for position estimation of the terminal 3.

  Here, the movable object as the recognition target according to the “object recognition information” can be recognized by an object recognition unit such as a person, an animal, a vehicle, or other movable physical object (for example, the camera 4). A variety of things fall under this category (if it can be taken by Furthermore, the shooting location is not particularly limited. For example, in addition to the inside of a store, it may be indoors such as a company, a school, a hall, a house, and a spectator or commuter. It may be outdoors where shoppers, workers, pedestrians, runners, etc. can be reflected.

  As yet another embodiment, the terminal 3 includes the above-described configurations (A) to (C), and functions as a terminal position estimation device that estimates its own position and further the positions of other terminals 3 around it. May be. In this case, for example, the “object recognition information” can be generated using a camera built in the terminal 3 (position is determined every moment).

Further, an invention (electromagnetic wave source position estimating apparatus 1 ′) closely related to the invention of the terminal position estimating apparatus 1 described above will be described. This electromagnetic wave source position estimation device 1 ′ acquires “received electromagnetic wave information” in a receiving device capable of receiving electromagnetic waves (radio waves, infrared rays, etc.) from a movable electromagnetic wave source 3 ′, and the position or position distribution of the electromagnetic wave source A device for estimating
(A ′) It is possible to acquire and receive “object recognition information” from a holding body (for example, a person) that can move while holding the electromagnetic wave source 3 ′ or from an object recognition means that can recognize the electromagnetic wave source 3 ′ itself. Feature data including “received electromagnetic wave information” at a certain time of the device and “object recognition information” at the time, and “electromagnetic wave source position or position distribution related to the position or position distribution of the electromagnetic wave source 3 ′ at the time Learning data generation unit 114 ′ that generates learning data including “information”;
(B ′) a model generation unit 115 ′ that generates a “model” for electromagnetic wave source position or position distribution estimation using a plurality of learning data;
(C ′) Using the generated “model”, based on feature data including “received electromagnetic wave information” at the position or position distribution estimation time point and “object recognition information” at the position or position distribution estimation time point, And an electromagnetic wave source position determination unit 116 ′ that determines position or position distribution information related to the electromagnetic wave source 3 ′ at the time of estimation of the position or position distribution.

  According to such an electromagnetic wave source position estimation apparatus 1 ′, in a situation where there is a movable electromagnetic wave source 3 ′ (a person carrying it), the calculation of the distance between the electromagnetic wave source 3 ′ and the receiving device is not performed. The position or position distribution of the electromagnetic wave source 3 ′ can be estimated with higher accuracy.

  For example, in a factory where a camera 4 'is installed in a factory, all workers carry out work with beacons (electromagnetic wave sources 3'), while receivers as receiving devices are installed at a plurality of locations in the workplace. When the electromagnetic wave source position estimation device 1 'is, the distribution state of each worker at the work place is grasped every moment from the received electromagnetic wave information at the installed receiver and the object identification information by the camera 4'. Is also possible.

  Incidentally, the object recognition means, the learning data generation unit 114 ′, the model generation unit 115 ′, the “model”, and the terminal position determination unit 116 ′ in the electromagnetic wave source position estimation apparatus 1 ′ are respectively This also corresponds to a functional component or component of the same name or corresponding name. As a further modification, in such an electromagnetic wave source position estimating apparatus 1 ′ and the terminal position estimating apparatus 1 described above, the same position estimation may be performed using ultrasonic waves (inaudible sound) instead of electromagnetic waves. It is possible.

[Configuration and function of terminal location estimation device]
FIG. 2 is a functional block diagram showing a functional configuration in an embodiment of the terminal position estimation apparatus according to the present invention.

  According to FIG. 2, the terminal position estimation apparatus 1 includes a communication interface 101 that can be connected to the camera 4, the terminal 3, and the like, a received electromagnetic wave information storage unit 102, an object recognition information storage unit 103, and terminal flow line information storage. Unit 104, display keyboard (DP / KB) 105, and processor memory. Here, the processor memory stores an embodiment of the terminal position estimation program according to the present invention, and has a computer function. By executing the terminal position estimation program, the terminal position estimation process is performed. To implement.

  Further, the processor memory includes, as function components, an object recognition unit 111, an object recognition information management unit 112, a received electromagnetic wave information management unit 113, a learning data generation unit 114, a model generation unit 115, and an object position comparison. A terminal location determination unit 116 including a unit 116a, a terminal flow line information management unit 117, an input / output control unit 118, and a communication control unit 121. These functional components can be regarded as functions of the terminal location estimation program stored in the processor memory, and the functional components of the terminal location estimation device 1 in FIG. 2 are connected by arrows. The process flow is also understood as an embodiment of the terminal location estimation method according to the present invention.

  Similarly, in FIG. 2, the camera 4 is a visible light, near-infrared, or infrared-capable imaging device including a solid-state imaging device such as a CCD image sensor or a CMOS image sensor. The camera 4 or a camera control device (not shown) generates captured image data including an image of an object captured by the camera 4 and transmits the information to the terminal position estimation device 1 in time series or batch. It has the function to do.

  The communication interface 101 receives captured image data that is a time-series image group from the camera 4 or the camera control device via a wired or wireless communication network. The acquired captured image data (image file) is output to the object recognition unit 111. Further, the communication interface 101 receives received electromagnetic wave information from the terminal 3 directly (or via a relay device such as an access point) via the wireless communication network. Alternatively, the received electromagnetic wave information may be received after being connected to the terminal 3 via a USB (Universal Serial Bus) or the like.

  The communication interface 101 may further transmit the flow line information generated as a result of the terminal position estimation to an external information processing apparatus. In addition, transmission / reception of such information in the communication interface 101 is controlled by the communication control unit 121, and information is exchanged at an appropriate timing.

  In this embodiment, the object recognizing unit 111 recognizes a person who stays or moves in the store, which is the position estimation area, based on the captured image data related to the image captured by the camera 4, and “object (person) recognition information”. Is generated. Here, the generated “object (person) recognition information” may include a two-dimensional (or three-dimensional) coordinate value of the foot position of one or more recognized persons at each time of recognition processing. it can. Note that the object (person) recognition process in the object recognition unit 111 can be performed by employing various known methods.

  For example, when an omnidirectional camera is used as the camera 4, Tatsuya Kobayashi, Haruhisa Kato, Masaru Sugano, “Proposal of a person flow tracking technique robust to occlusion using a monocular omnidirectional camera”, Shingaku It is possible to generate object (person) recognition information using the technique described in Technical Bulletin, vol.116, no.411, PRMU2016-146, 2017, pages 321-326.

  Specifically, in this method, person detection processing and person tracking processing are executed for each image frame, the position of each person at each time is estimated, and the person is tracked. Among these, in the person detection process, matching is performed between a region extracted using the background difference technique and a 3D human model prepared in advance to detect a human region in the image. Here, a person model is arranged in advance at various foot positions where a person can exist, and the corresponding foot position is detected as a person position by searching for a contour having the highest similarity to the foreground region.

  Further, in the person tracking process, the person ID being tracked is assigned to the detected person area by associating the person being tracked with the above detection result, and the element having the maximum matching score by one-to-one person matching is determined. select. Next, based on the result of this person matching, the tracking position of each person is estimated using a particle filter.

  Incidentally, in the present apparatus 1, the object recognition unit 111 is not provided, and the object recognition information management unit 112 may take an embodiment in which object recognition information is acquired and managed from an external object recognition unit. In any case, the object recognition information management unit 112 manages the acquired “object recognition information” while being stored in the object identification information storage unit 103, and further performs position estimation to the learning data generation unit 114 in the learning phase. In the phase, the information is appropriately output to the terminal position determination unit 116.

  The received electromagnetic wave information management unit 113 manages the “received electromagnetic wave information” acquired from the terminal 3 while saving it in the received electromagnetic wave information storage unit 102, and in the learning phase, the learning data generation unit 114 further performs the position estimation phase. Then, the data is output to the terminal position determination unit 116 as appropriate. Here, the “received electromagnetic wave information” is information in which the reception time, RSSI, and ID of the beacon 2 are associated with each other, or a part of the information for each beacon 2 that is the source of the electromagnetic wave received by the terminal 3. be able to.

The learning data generation unit 114 uses the acquired “object recognition information” and “received electromagnetic wave information”,
(A) Feature data including “received electromagnetic wave information” at a certain time of the terminal 3 and “object recognition information” (object position information) at the time;
(B) Generate “learning data” including “terminal position information” (as correct data) related to the position of the terminal 3 at that time.

  FIG. 3 is a schematic diagram schematically showing a terminal location estimation process according to the present invention including a learning phase, a location estimation phase, and a model update phase. As one embodiment, the learning data generation unit 114 generates feature data as shown in <Learning Phase> of FIG. 3 and correct terminal position data (terminal position information) as learning data.

Specifically, the feature data shown in FIG. 3 includes (a) RSSI_i (i = 1, 2,..., I at a certain point in time at each of a large number of positions (position coordinates (x _p , y _p )). ): RSSI actual measurement value of electromagnetic wave received from beacon i, and (b) obstacle_j (j = 1, 2,..., J): of a movable object (for example, a person) derived from a captured image of the camera 4 Contains the position coordinate value. In FIG. 3, each row in the feature data table is one feature data.

Here, as the above-mentioned many positions (position coordinates (x _p , y _p )), for example, after a terminal position estimation area such as a floor in a store is divided by a mesh, a representative position of the mesh (for example, a mesh position) It is also preferable to adopt (center). In this case, the RSSI_i in (a) is also a value measured by actually placing the terminal 3 at the mesh representative position. For RSSI_i, when there are a plurality of RSSI actual measurement values at the position, it is also preferable to adopt an average value or a median value thereof.

  Further, the number (J) of obstacle_j in (b) in one feature data is preset to the upper limit of the number of movable objects (persons) that can be included in the captured image of the camera 4 or a value exceeding the upper limit. can do. If the number of objects (persons) identified from the captured image of the camera 4 is J ′ (J ′ <J), for example, each value from obstacle_1 to obstacle_J ′ is a derived position coordinate value, and obstacle_J ′ Any of +1 to obstacle_J may be zero (however, in this case, the position coordinate is set so that the coordinate value does not take a zero value). Alternatively, as each value from obstacle_J ′ + 1 to obstacle_J, a predetermined value that cannot be taken as a position coordinate value may be set.

  Further, as a modification, an ID (for example, id = 1, 2,..., J ′) is assigned to a movable object (person) captured by the hand from the photographed image of the camera 4, for example, from obstacle_1 to obstacle_J ′. Each may be a position coordinate value of the movable object (person) specified by the ID.

Returning to FIG. 2, the model generation unit 115 uses the “learning data” generated by the learning data generation unit 114 to generate a position estimation model for terminal position estimation. As this position estimation model,
Either (a) a position estimation discriminator constructed by machine learning of “learning data”, or (b) a terminal position database constituted by “learning data”.

  Here, the position estimation model according to the machine learning of (a) may be generated by, for example, Xgboost (eXtreme gradient boosting), random forest (Random Forest), or neural network (Neural Network). Among them, Xgboost uses samples extracted at random like a random forest, but it is easy to deal with feature quantities of continuous variables, and it integrates multiple learning results derived through repeated weight adjustments to estimate accuracy. Is a machine learning technique that can generate a model that is robust against outliers and missing values.

  On the other hand, the terminal location database of (b) is, for example, table information in which corresponding feature data (at the terminal location) is recorded in association with each piece of terminal location information (correct terminal location data in FIG. 3). be able to. For example, it may be table data in which {RSSI_i} data and {obstacle_j} data at the representative position are recorded in association with each representative position of the mesh dividing the terminal position estimation region at each time of the database generation period. .

  In any case, when the model generation unit 115 generates a position estimation model such as the above (a) or (b), for example, in the case of a store, the model generation unit 115 may detect an electromagnetic shielding object such as a desk or a shelf installed in the store. There is no need for specific layout information, and there is no need to incorporate specific arrangement information of the installed beacons 2. Furthermore, since the learning phase is advanced based on the actual measured value of RSSI, for example, without calculating the three-dimensional distance between the electromagnetic wave source and the terminal or the three-dimensional size of the electromagnetic wave shielding object such as a desk or shelf It is possible to generate a position estimation model that includes a three-dimensional effect (for example, the degree of shielding by height) of an installed electromagnetic shielding object or a movable object (for example, a person). As a result, higher position estimation accuracy is realized.

  Similarly, in FIG. 2, the terminal position determination unit 116 uses the position estimation model generated as described above, “reception electromagnetic wave information” at the time of position estimation acquired from the terminal 3 as the position estimation target, and the position estimation. Based on the feature data including the “object recognition information” (object position information) at the time, the terminal position information related to the position estimation target terminal 3 is determined.

  Specifically, when the position estimation model is the position estimation discriminator of the above (a), the terminal position determination unit 116, as one embodiment, performs a time T (as shown in <position estimation phase> in FIG. The position estimation target acquisition data (feature data) at the position estimation target terminal 3 at t = T) is input to the position estimation model, and the output from this position estimation model is acquired as the estimated position of the terminal 3 at time T To do.

Here, the position estimation time acquisition data (feature data) shown in FIG. 3 is the time of the position estimation period (t = T, T-1, T-2,... (Data generation time unit for simplification). 1)))
(A) RSSI_i (i = 1, 2,..., I) at the terminal 3 as a position estimation target: RSSI measured value of the electromagnetic wave received from the beacon i, and (b) obstacle_j (j = 1, 2, ..., J): The position coordinate value of the movable object (for example, a person) derived from the captured image of the camera 4 is included. In FIG. 3, each row in the position estimation time acquisition data table is one position estimation time acquisition data. (Characteristic data).

  The position estimation model inputs such position-acquisition time acquisition data, and “estimates” as terminal position estimation values at each time (t = T, T-1, T-2,...) During the position estimation period. The “position coordinates” and the “reliability” (probability) in the position coordinate values are output. Here, the “reliability” of the “estimated position coordinates” is the highest value among the output position estimated values.

  On the other hand, when the position estimation model is the terminal position database of (b) above, the terminal position determination unit 116 compares the position estimation time acquisition data (feature data) as described above with each record of the terminal position database. (Similarity between the two is calculated), the most similar record is determined, and the position coordinates of the record are determined as the terminal position estimated value.

Specifically, in the record (row) p of the terminal location database,
(A) The position coordinates are x _p , y _p ,
(B) Let {RSSI_i} be f _{p, 1} , f _{p, 2} ,..., F _{p, I} ,
(C) Let {obstacle_j} be f _{p, I + 1} , f _{p, I + 2} ,..., F _{p, I + J}
On the other hand, at time T of the position estimation time acquisition data,
(D) Let {RSSI_i} be gT _{, 1} , gT _{, 2} , ..., gT _{, I} ,
(E) Let {obstacle_j} be g _{T, I + 1} , g _{T, I + 2} ,..., G _{T, I + J.}

In this case, the distance between the position estimation time acquisition data at time T and the record p, that is, the similarity dT_p is expressed by the following equation (1) dT_p = ((g _{T, 1} −f _{p, 1} ) ² + (g _{T, 2} −f _{p, 2} ) ² + ・ ・ ・ + (g _{T, I + J} −f _{p, I + J} ) ² ) ^0.5
Is calculated using Among the records in the terminal position database, the position coordinates (x _p , y _p ) of the above (a) in the record having the smallest similarity dT_p is determined as the terminal position estimated value at time T. When this "estimated position coordinate" (est_x, est_y) is described in a programming language, for example,
(2) est_x = database [argmin {dT_p}] [x _p ]
est_y = database [argmin {dT_p}] [y _p ]
It becomes.

  Here, the similarity (distance) dT_p of the “estimated position coordinates” is the anti-reliability in the estimated position coordinates. For example, a monotonically decreasing function of the distance dT_p such as the reciprocal 1 / dT_p of the distance is “reliable” Degree ".

  Returning to FIG. 2, another embodiment of the terminal position estimation process will be described next. In the present embodiment, a situation is considered in which a plurality of “terminal position information”, which are estimated position candidates, are output (determined) from the position estimation model. For example, among the “terminal position information” output (determined) as an estimated position from the position estimation model, information having a reliability equal to or higher than a predetermined threshold can be selected as an estimated position candidate.

  In this case, the object position comparison unit 116a of the terminal position determination unit 116 compares the “object position information” at the time of position estimation with “terminal position information” that is an estimated position candidate derived from the position estimation model. Then, the closeness (that is, distance) of the positions of each other is calculated. The terminal position determination unit 116 estimates the terminal position of the position estimation target terminal 3 based on the closeness (small distance) of the mutual positions.

  There is a possibility that the position estimation target terminal 3 is shown in a photographed image by the camera 4 and is present at a position related to any of the “object position information” derived from the photographed image or a position in the vicinity thereof. Very expensive. Therefore, by obtaining the estimated position of the position estimation target based on the closeness (small distance) of the position of the estimated position candidate “terminal position information” and “object position information”, more accurate position estimation Is possible.

  More specifically, the terminal position determination unit 116 includes, among the “object position information” at the time of position estimation, a position related to the “object position information” and a determination terminal “terminal” derived from the position estimation model The “object position information” having the shortest distance from the position related to “position information” or less than or equal to a predetermined threshold value may be determined as the estimated position related to the terminal 3 as the position estimation target.

  Further, as a change mode, the terminal position determination unit 116 includes the position related to the “terminal position information” among the “terminal position information” that is the determination candidate derived from the position estimation model, and the “object” at the position estimation time point. It is also preferable to determine “terminal position information” having the shortest distance from the position related to “position information” or a predetermined threshold value or less as the estimated position related to the terminal 3 as the position estimation target.

  Further, as a further change mode, the terminal position determination unit 116 selects only “terminal position information” whose reliability is greater than or equal to a predetermined threshold among the “terminal position information” of the estimated position candidates derived from the position estimation model. Compared with the “object position information” at the time of position estimation, the estimated position related to the position estimation target terminal 3 may be determined.

  Similarly, in FIG. 2, the terminal flow line information management unit 117 collects the estimated position coordinates of the terminal 3 at each time during the position estimation period determined by the terminal position determination unit 116, and the terminal 3 (the user who owns the terminal 3). For each identifier ID, flow line information in which time series information of the collected estimated position coordinates is associated is generated.

  The terminal flow line information management unit 117 may store the generated flow line information in the terminal flow line information storage unit 104, and preferably displays the generated flow line information on the display 105 via the input / output control unit 118. You may transmit to an external information processing apparatus via the control part 121 and the communication interface 101. FIG. Note that the processing of the flow line information may be appropriately executed as set in advance, or may be executed by receiving a processing command from the user via the keyboard 105, for example.

  Next, <model update phase> in the terminal position estimation apparatus 1 will be described with reference to FIG.

  As shown in <model update phase> in FIG. 3, the terminal position determination unit 116 determines, from the position estimation model, “terminal position information” and its reliability (probability, distance 1 / dT_p) for each time in the position estimation period. And the “terminal position information” is determined as the estimated position.

  Here, when the “terminal position information”, which is the determined estimated position, has a reliability R that is equal to or greater than a predetermined threshold ThR (R ≧ ThR), the learning data generation unit 114 sets the “terminal position information” as a correct answer. Learning data included as terminal location information is generated.

  Next, the model generation unit 115 (FIG. 2) updates the position estimation model (position estimation discriminator, terminal position database) using this learning data. Thereby, for example, the position of the terminal 3 can be learned online, and the position can be estimated almost in real time with higher accuracy. A new position estimation model may be constructed using such learning data.

[Other Embodiments of Terminal Position Estimation Processing]
Next, an embodiment of the terminal position estimation process in consideration of the position at the time before the position estimation time in the terminal 3 and also the reliability will be described.

The learning data generation unit 114, as one embodiment,
(A) “terminal position information” relating to the position at a time point (eg, t1-1) before a certain time point (t1) in the terminal 3, and (b) the previous time point (eg, t1-1) in the terminal 3. "Received electromagnetic wave information"
The learning data further including both or any one of these is generated, and the model generation unit 115 generates a position estimation model using the learning data.

Next, the terminal position determination unit 116 matches the above learning data,
(A) terminal position information relating to a position at a time point before the position estimation time point (T) (for example, T−1) in the position estimation target terminal 3; and (b) the previous position information in the position estimation target terminal 3 "Received electromagnetic wave information" at the time (eg T-1)
The terminal position information of the terminal 3 as the position estimation target is determined based on the feature data that further includes both or one of the above.

  Normally, the position of the terminal 3 (the person who owns it) shows a correlation with the position at the previous time point, and the path in which the person can move is installed objects such as desks and shelves, particularly in a store. It is recognized that both have a strong correlation when substantially limited by Therefore, as in this embodiment, the position estimation accuracy can be further improved by learning or creating a database of the position at the previous time point.

Further, as another embodiment using the position at the previous time point, the terminal position determination unit 116 has “terminal position information” having a reliability R (≧ ThR) equal to or higher than a predetermined threshold ThR at the position estimation time point (T). Is obtained multiple times,
(A) “terminal location information” closest to the estimated location at the previous time point (eg, T−1),
(B) “terminal position information” whose distance from the estimated position at the previous time point (eg, T−1) is less than a predetermined threshold, or (c) the previous time point (eg, T−1) for the terminal 3 that is the position estimation target. ) To predict the position at the position estimation time point (T) (for example, assuming that the position change vector has been displaced by the average of the past time points) and the closest to the predicted position or “Terminal location information” whose distance from the predicted location is less than a predetermined threshold
Is also preferably determined to be an estimated position.

  Furthermore, as an embodiment using the reliability at the previous time point, the learning data generation unit 114 includes a time point before the time point (t1) related to the “terminal position information” included in the generated learning data (for example, t1-1). The learning data further including the reliability calculated for the “terminal position information” according to () may be generated.

  In this case, the terminal position determination unit 116 calculates the reliability calculated for the “terminal position information” related to the position at the time (eg, T−1) before the position estimation time (T) in the terminal 3 that is the position estimation target. The “terminal position information” relating to the position estimation target terminal 3 can also be determined based on the degree. As described above, the position estimation accuracy can be further improved by using the position estimation model considering the reliability.

As yet another embodiment, the terminal position determination unit 116, when there are a plurality of “terminal position information” (as estimated position candidates) having a reliability R (≧ ThR) equal to or greater than a predetermined threshold ThR,
(A) For “terminal position information” determined at a time point before the position estimation time point (T) (for example, T−1), or (b) for a past time point of the position estimation time point (T) (for example, T−1, For “terminal position information” predicted from the “terminal position information” group (trajectory information) determined in T−2,.
It is also preferable to select “terminal position information” having close proximity to a position that satisfies a predetermined condition, and to determine “terminal position information” related to the terminal 3 as a position estimation target.

[Terminal location estimation method]
FIG. 4 is a flowchart schematically showing an embodiment of a location estimation phase in the terminal location estimation method according to the present invention. According to this flowchart, first, steps S11 to S14 are executed at each time (t = 1, 2,..., T−2, T−1, T) in the position estimation period.

(S11) The RSSI is acquired from the position estimation target terminal 3, and the camera image is further acquired to determine the position coordinates of the dynamic object (for example, a person).
(S12) Feature data at the position estimation time is generated from the synchronized RSSI and the position coordinates of the dynamic object.

(S13) Based on the generated feature data, the estimated position coordinates of the terminal 3 as the position estimation target are determined using the position estimation model generated in advance.
(S14) The flow line information (time series information of estimated position coordinates) of the position estimation target terminal is generated / updated.

Next, after the end of the position estimation period, the following step S15 is executed, and the position estimation process in the position estimation period ends.
(S15) The generated flow line information of the terminal 3 is output.

  By executing the processes in steps S11 to S15 described above for each terminal 3 in the position estimation area, it is possible to grasp the flow lines of all the terminals 3. In the present embodiment, the RSSI in step S11 and the position coordinates of the dynamic object are synchronized, and feature data suitable for the situation at the time is generated at each time. As a result, a position with higher accuracy is obtained. Estimation is possible.

  As described above in detail, according to the present invention, a model is generated based on object recognition information including information on a movable object, and position estimation is performed using such a model. Even in a situation where there can be a movable object that affects the electromagnetic wave from the terminal, the position of the terminal that has received the electromagnetic wave can be estimated with higher accuracy.

  By the way, the configuration and method of the present invention can be applied to, for example, a monitoring system for monitoring a place where a large number of persons move, stay, and go in and out, as well as in one store, in a commercial / service facility, and further in a store. The present invention can be applied to various systems such as a marketing research system for investigating the entrance / exit of a person on a city street, a break, watching a game, participating in an event, and the situation of movement.

  In the various embodiments of the present invention described above, various changes, modifications, and omissions in the technical idea and scope of the present invention can be easily made by those skilled in the art. The above description is merely an example, and is not intended to be restrictive. The invention is limited only as defined in the following claims and the equivalents thereto.

DESCRIPTION OF SYMBOLS 1 Terminal position estimation apparatus 101 Communication interface 102 Received electromagnetic wave information storage part 103 Object recognition information storage part 104 Terminal flow line information storage part 105 Display keyboard (DP / KB)
DESCRIPTION OF SYMBOLS 111 Object recognition part 112 Object recognition information management part 113 Received electromagnetic wave information management part 114 Learning data generation part 115 Model generation part 116 Terminal position determination part 116a Object position comparison part 117 Terminal flow line information management part 118 Input / output control part 121 Communication control Part 2 Beacon (electromagnetic wave source)
3 Terminal 4 Camera

Claims (15)

A terminal position estimation device that obtains received electromagnetic wave information in a terminal that has received electromagnetic waves from an electromagnetic wave source and estimates the position of the terminal,
Object recognition information from an object recognition means capable of recognizing a movable object can be acquired, the received electromagnetic wave information at a certain time of the terminal, and feature data including the object recognition information at the time, Learning data generating means for generating learning data including terminal position information relating to the position of the terminal at the time,
Model generation means for generating a model for terminal position estimation using a plurality of the learning data;
Based on the feature data including the received electromagnetic wave information at the position estimation time acquired from the position estimation target terminal and the object recognition information at the position estimation time using the model, the position estimation target terminal A terminal position determining means for determining the terminal position information. The object identification information is object position information related to the position of the recognized object,
The terminal position determination means compares the object position information at the position estimation time with the terminal position information that is a determination candidate derived from the model, and determines the position based on the proximity of each other position. The terminal location estimation apparatus according to claim 1, wherein terminal location information relating to a terminal to be estimated is determined. The terminal position determination means is a distance between a position related to the object position information and a position related to the terminal position information that is a candidate for determination derived from the model among the object position information at the time of the position estimation. 3. The terminal position estimation apparatus according to claim 2, wherein object position information having the smallest value or less than a predetermined threshold is determined as terminal position information related to the position estimation target terminal. The terminal position determination means is a distance between a position related to the terminal position information and a position related to the object position information at the time of position estimation among the terminal position information that is a candidate for determination derived from the model. 3. The terminal position estimation apparatus according to claim 2, wherein terminal position information having a smallest value or less than a predetermined threshold is determined as terminal position information related to the position estimation target terminal.   The terminal position determination means calculates a determination reliability for each terminal position information that is a determination candidate derived from the model, and determines only the terminal position information whose reliability is equal to or greater than a predetermined threshold. 5. The terminal position estimation apparatus according to claim 2, wherein terminal position information relating to the position estimation target terminal is determined in comparison with the object position information at the time. The terminal position determination means calculates a reliability of determination for each terminal position information that is a determination candidate derived from the model,
The learning data generation means generates the learning data including the terminal position information as correct terminal position information when the determined terminal position information has the reliability equal to or higher than a predetermined threshold.
The terminal location estimation apparatus according to claim 1, wherein the model generation unit generates or updates the model using the learning data. The learning data generating means includes the learning data further including terminal position information relating to a position of the terminal at the time before the time and / or the received electromagnetic wave information at a time before the time of the terminal. Generate
The terminal position determination means is the terminal position information related to the position at the time before the position estimation time at the position estimation target terminal and / or the time before the position estimation time at the position estimation target terminal. 7. The terminal location estimation apparatus according to claim 1, wherein terminal location information related to the location estimation target terminal is determined based on feature data further including the received electromagnetic wave information. . The terminal position determination means calculates a reliability of determination for each terminal position information that is a determination candidate derived from the model,
The learning data generation means generates the learning data further including the reliability calculated for the terminal position information related to the time point before the time point related to the terminal position information included in the generated learning data,
The terminal position determination means is based on the reliability calculated for the terminal position information related to the position at the time before the position estimation time in the position estimation target terminal, based on the reliability. The terminal position estimation apparatus according to any one of claims 1 to 7, wherein the terminal position information according to claim 1 is determined.   The terminal position determination means calculates a determination reliability for each terminal position information that is a determination candidate derived from the model, and when there are a plurality of the terminal position information having the reliability equal to or higher than a predetermined threshold. The position of the position satisfying the predetermined condition with respect to the terminal position information determined at the time before the position estimation time or with respect to the terminal position information predicted from the terminal position information group determined at the past time of the position estimation time The terminal position estimation apparatus according to any one of claims 1 to 8, wherein the terminal position information having closeness is selected and determined as terminal position information related to the terminal of the position estimation target.   The terminal position estimation apparatus according to claim 1, wherein the model generation unit generates a terminal position estimation model by machine learning as the model.   The terminal position estimation apparatus according to claim 1, wherein the model generation unit generates a terminal position database generated by a plurality of the learning data as the model.   The object recognition information is generated based on photographing or measurement information from at least one of a camera, an omnidirectional camera, a depth camera, an infrared camera, an infrared positioning unit, a laser positioning unit, and a thermographic unit. The terminal position estimation apparatus according to any one of claims 1 to 11. An electromagnetic wave source position estimating device that acquires received electromagnetic wave information in a receiving device capable of receiving an electromagnetic wave from a movable electromagnetic wave source and estimates a position or position distribution of the electromagnetic wave source,
It is possible to acquire the object recognition information from the holding body that can move while holding the electromagnetic wave source, or the object recognition means that can recognize the electromagnetic wave source, the received electromagnetic wave information at a certain point of the receiving device, and Learning data generating means for generating learning data including feature data including the object recognition information at the time and electromagnetic wave source position or position distribution information related to the position or position distribution of the electromagnetic wave source at the time;
Model generation means for generating a model for electromagnetic wave source position or position distribution estimation using a plurality of the learning data,
Using the model, based on feature data including the received electromagnetic wave information at the position or position distribution estimation time and the object recognition information at the position or position distribution estimation time, the position or position distribution estimation time An electromagnetic wave source position estimating device comprising: electromagnetic wave source position determining means for determining position or position distribution information relating to the electromagnetic wave source. A program that causes a computer mounted on a device that acquires received electromagnetic wave information in a terminal that receives an electromagnetic wave from an electromagnetic wave source and estimates the position of the terminal to function,
Object recognition information from an object recognition means capable of recognizing a movable object can be acquired, the received electromagnetic wave information at a certain time of the terminal, and feature data including the object recognition information at the time, Learning data generating means for generating learning data including terminal position information relating to the position of the terminal at the time,
Model generation means for generating a model for terminal position estimation using a plurality of the learning data;
Based on the feature data including the received electromagnetic wave information at the position estimation time acquired from the position estimation target terminal and the object recognition information at the position estimation time using the model, the position estimation target terminal A terminal position estimation program for causing a computer to function as terminal position determination means for determining such terminal position information. A terminal position estimation method by a computer mounted on a device that acquires received electromagnetic wave information in a terminal that has received electromagnetic waves from an electromagnetic wave source and estimates the position of the terminal,
Object recognition information from an object recognition means capable of recognizing a movable object can be acquired, the received electromagnetic wave information at a certain time of the terminal, and feature data including the object recognition information at the time, Generating learning data including terminal position information related to the position of the terminal at the time,
Generating a model for terminal position estimation using a plurality of the learning data;
Based on the feature data including the received electromagnetic wave information at the position estimation time acquired from the position estimation target terminal and the object recognition information at the position estimation time using the model, the position estimation target terminal And determining the terminal position information.

Images