JP2019027909A - Method for estimating position of target by unmanned mobile body - Google Patents

Abstract

To provide a technique for actively estimating the position of a missing person by estimating the distribution of the position of the missing person and updating the distribution every time an observation is obtained.SOLUTION: The method for estimating a position according to the present invention is executed by an unmanned mobile body. The method includes the steps of: temporarily determining the estimated position of a target; observing a signal sent from the target by the unmanned mobile body; updating the estimated position and then determining the next observation place; moving to the observation place after the update and making an observation again. The method repeats the series of above steps to gradually estimate the position of the target.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a method for estimating a position of an object by an unmanned moving body.

  In recent years, the development of UAV (Unmanned Aerial Vehicle) has been remarkable, and its uses such as aerial photography and search are various. There are some studies to search for targets by UAV, but the way the drone moves is predetermined.

  In addition, when searching for missing persons in mountains and forests, it is important to find missing persons as quickly as possible, but to narrow down the position only from the last sighting information etc. of missing persons. It is difficult, and it is not uncommon for an exhaustive search to be finally performed as in the roller operation.

  Consider climbing a person who has a beacon in advance, and if he is missing, estimate the position based on the signal from the beacon and perform an efficient search. There are several methods for detecting the position of a signal source by observing signals with a plurality of sensors or performing observations with a single sensor multiple times. These methods can be broadly classified into two groups.

  One is a method called a range-based method in which the arrival direction of a received signal and the distance to the signal source are calculated and the position is detected based on the information. The other is a method that performs position detection without calculating the distance from the received signal to the signal source, and is called a range-free method. In general, the range-based method has higher accuracy but higher calculation cost. Conversely, the range-free method has lower accuracy but lower calculation cost and is robust.

Tomoaki Ohtsuki. Position estimation technology. IEICE Technical Report (2009)

  The technology described in Non-Patent Document 1 has little to do with appropriate selection of observation locations.

  Therefore, the present invention proposes a method for actively estimating the position of the victim by estimating the distribution of the position of the victim using the particle filter and updating the distribution every time the observation is obtained. In addition, we show that it is possible to perform highly accurate estimation with a small number of observations by selecting an appropriate observation location. One object is to provide a technique for quickly collecting a wide range of information.

According to the present invention, there is provided a method for estimating the position of an object by an unmanned moving body,
A step of tentatively determining an estimated position of the object; a step of observing a signal transmitted from the object by the unmanned mobile body; a step of updating the estimated position and determining a next observation place after the update; Moving to a later observation location and performing the observation again, and gradually estimating the position of the object,
A position estimation method is obtained.

  According to the present invention, a wide range of information can be collected quickly.

It is a functional block diagram of the unmanned air vehicle according to the embodiment of the present invention. It is a functional block diagram of a management server. It is a simulation result of the estimation method by this Embodiment.

The contents of the embodiment of the present invention will be listed and described. A position estimation method for an object by an unmanned moving object according to an embodiment of the present invention has the following configuration.
[Item 1]
A method for estimating the position of an object by an unmanned moving object,
A step of tentatively determining an estimated position of the object; a step of observing a signal transmitted from the object by the unmanned mobile body; a step of updating the estimated position and determining a next observation place after the update; Moving to a later observation location and performing the observation again, and gradually estimating the position of the object,
Position estimation method.
[Item 2]
The position estimation method according to claim 1,
The step of observing the signal uses a particle filter, and the step of determining the next observation location uses active learning.
Position estimation method.

<Details of the embodiment>
Hereinafter, a position estimation system using a method for estimating the position of an object by an unmanned moving object according to an embodiment of the present invention will be described.

<Overview>
The position estimation system according to the embodiment of the present invention can be realized using an unmanned air vehicle and a computer terminal.

  The unmanned aerial vehicle in the present embodiment is a drone, a multicopter, an unmanned aerial vehicle (UAV), a RPAS (remote piloted air systems), or a UAS (Amant lands). May be. The unmanned air vehicle includes a battery, a plurality of motors, a position detection unit, a control unit, a driver, a storage device, a wireless communication device, a voltage sensor, a current sensor, and the like. These components are mounted on a frame having a predetermined shape. The hardware configuration of the information processing apparatus mounted on the unmanned air vehicle will be described later. As for the basic structure for these flights, a known technique can be appropriately adopted.

<Unmanned flying vehicle>
As shown in FIG. 2, the unmanned air vehicle 1 includes an information processing apparatus 100. The information processing apparatus 100 configures a part of the information transmission system by executing information processing via communication with the server 1. The information processing apparatus 100 includes at least a processor 10, a memory 11, a storage 12, a transmission / reception unit 13, an output unit 14, a positioning unit 16, a detection unit 17, and the like, which are electrically connected to each other through a bus 15. The information processing apparatus 100 may be configured by, for example, a microcomputer, an ASIC (Application Specific Integrated Circuit), or may be logically realized by cloud computing.

  The processor 10 is an arithmetic device that controls the operation of the information processing apparatus 100, performs control of data transmission / reception between elements, processing necessary for execution of an application, and the like. For example, the processor 10 is a CPU and / or GPU (Graphical Processing Unit) or the like, and executes each necessary information process by executing a program stored in the storage 12 and expanded in the memory 11.

  The memory 11 includes a main memory composed of a volatile storage device such as a RAM and an auxiliary memory composed of a nonvolatile storage device such as a flash memory or an HDD. The memory 11 is used as a work area of the processor 10 and stores a BIOS executed when the information processing apparatus 100 is started up, various setting information, and the like. The storage 12 stores application programs and the like.

  The transmission / reception unit 13 connects the information processing apparatus 100 to the network 4 and communicates with the management server 1 via the LPEA network. The transmission / reception unit 13 may include a Bluetooth (registered trademark) and a BLE (Bluetooth Low Energy) short-range communication interface.

  The input / output unit 14 is an information input device such as switches and an output device such as a display. The unmanned air vehicle 1 performs autonomous flight, but may be operated manually or automatically remotely from the outside. The unmanned aerial vehicle 1 according to the present embodiment includes a camera as an input function, and can take aerial photographs of still images and moving images. In addition, various cameras such as an infrared thermo camera, an X-ray camera, a high sensitivity camera, and a night vision camera may be provided according to information to be collected.

  The bus 15 is commonly connected to the above-described elements, and transmits, for example, an address signal, a data signal, and various control signals.

  The positioning unit 16 detects at least the position and altitude of the unmanned air vehicle 1. The positioning unit 26 according to the present embodiment is, for example, a GPS (Global Positioning System) detector, and detects the latitude, longitude, and altitude of the current position of the unmanned air vehicle 1.

  The detection unit 17 is for sensing the external environment of the unmanned air vehicle 1 with various sensors such as voice, image, infrared, and the like, and manages an auxiliary function of independent flight.

  The unmanned air vehicle 1 according to the present embodiment includes, in addition to the information processing device 100, a power source, a motor connected to the rotor, and the information processing device 100 and the motor for moving and flying the unmanned air vehicle 1. At least a relay driver is further included. The information processing apparatus 100 controls a plurality of motors and uses a surveillance drone flight control (control of ascending, descending, horizontal movement, etc.) and a gyro (not shown) mounted on the unmanned air vehicle 1. Attitude control is also performed by controlling a plurality of motors. The driver drives the motor according to a control signal from the information processing apparatus 100. For example, the motor is a DC motor, and the driver is a variable voltage power supply circuit that applies a voltage specified by a control signal to the motor. The unmanned aerial vehicle 100 may have other elements not shown.

<Management server>
As shown in FIG. 2, the management server 2 is an information processing apparatus for providing a service through an information transmission system, and may be a general-purpose computer such as a workstation or a personal computer, or cloud computing. It may be logically realized by. As shown in FIG. 2, the server 2 includes a processor 20, a memory 21, a storage 22, a transmission / reception unit 23, an input / output unit 24, and the like, which are electrically connected to each other through a bus 25.

  The processor 20 is an arithmetic device that controls the overall operation of the server 2 and performs control of data transmission / reception between elements, information processing necessary for executing an application, and the like. For example, the processor 20 is a CPU (Central Processing Unit), and executes each information process by executing a program stored in the storage 22 and expanded in the memory 21.

  The memory 21 includes a main memory composed of a volatile storage device such as a DRAM (Dynamic Random Access Memory) and an auxiliary memory composed of a nonvolatile storage device such as a flash memory or an HDD (Hard Disc Drive). . The memory 21 is used as a work area of the processor 20 and stores a BIOS (Basic Input / Output System) executed when the server 2 is started up, various setting information, and the like. The storage 22 stores various programs such as an application program and an authentication program for each unmanned air vehicle 1. A database (location data, route data, etc., which will be described later) storing data used for each process may be constructed in the storage 22.

<Method according to the invention>
In the position estimation using RSSI in related research, the position of the sensor for observing the signal is arranged in a grid around the sensor.

  In this method, observation is performed at once and position estimation is performed using the obtained data. In the present invention, since it is considered that the estimation proceeds while observing the beacon signal possessed by the victim with the sensor mounted on the UAV, the distribution is estimated using the Particle Filter, and updated every time the observation is obtained. I will do it. Furthermore, since the present invention considers the position estimation of the victim, it is also important to perform the position estimation with high accuracy with a small number of observations. For this reason, it is necessary to select an observation place appropriately.

  In order to solve this problem in the present invention, it is considered to apply a method used in Active learning, which is one field of machine learning. Active learning is a learning method for efficiently learning with as few correct labels as possible for data when it takes cost and time to obtain correct labels for certain data. In the present invention, a method is proposed that realizes efficient and highly accurate position estimation by combining the methods used in the ParticleFilter and Active learning.

  In the proposed method, the estimated position is updated by observing the signal, and the next observation location is determined after the update. The estimation is gradually advanced by repeating the operation of performing observation again at that place and updating the estimated position.

  In the present invention, since the distribution of the position of the victim is considered to be a non-Gaussian distribution, a simple Kalman filter cannot be used. Here, for ease of implementation, a particle filter is used to estimate the position of the signal source, and a method used in active learning is applied to determine the next observation location.

  In this embodiment, it is assumed that a sampling impor-tance resampling algorithm is used as the particle filter. That is, every time observation data is obtained, the likelihood is calculated for each particle, and resampling is performed according to the likelihood to update and estimate the state of the particle.

  Active learning is an algorithm for efficiently learning with as few correct labels as possible in data when it takes cost and time to obtain correct labels for certain data in one field of machine learning. In Active learning, there are several approaches to determine what data to get next. In the present invention, in order to estimate the position of the missing person as soon as possible, it is necessary to appropriately select a place to be observed and proceed with the estimation efficiently, so the technique used in Active learning is referred to. In the following sections, specific application methods of the technique used in the particle filter and active learning according to the present invention will be described.

<Proposed method based on Range-Based method>
In the proposed method based on the range-based method, position detection is performed using RSSI even in the range-based method. If a certain observation is obtained, the likelihood is set for each particle filter particle according to the difference from the theoretical value of the signal intensity, the estimation is updated by resampling, and the next observation location is determined. Repeat the above operation. The specific update method and the method of determining the observation location are described below. In the future, the unit distance d ₀ = 1 in the theoretical formula (1) will be treated.

<Update distribution>
The number of particles and N _p, receives the (i-1) th observation represents the particles after being resampled by Equation 1.
Further, if the particle number is represented by a subscript j (j = 1, 2,..., Np), each particle has information on the position of the particle, the signal intensity at a unit distance, and the attenuation coefficient. It can be expressed by Equation 2.
Further, the i-th observation place is represented by Equation 3 and the observation value is represented by Equation 4.
Here, when the observed value (Formula 4) is obtained, the likelihood w for each particle is calculated by the following Formula 5 by assuming that the observed noise follows a normal distribution with mean 0 and variance 2. Can do.
If the likelihood w can be calculated, an updated particle (Formula 6) can be obtained by resampling according to this.

<Determination of observation location>
When the i-th observation is received and the resampled particle (Equation 6) is obtained, the next observation should be considered. For example, suppose that observation is performed at a certain position [x ₀ , y ₀ ] ^T. At this time, a set R of Np observation values obtained from each particle depends on [x ₀ , y ₀ ] ^T, and can be written as Equation 7.
Even if the observation is performed at a position where Np observation values R _j (j = 1, 2,..., Np) have the same value, the distribution after the re-sampling after receiving the observation at that time is re-sampled. It does not change much from the distribution before sampling. Conversely, the distribution changes before and after resampling are larger when observations are made at locations where different observations are possible. Such a concept is called uncertain sampling in Active learning, and the observation location can be determined according to the following Equation 11.

<Simulation results>
The result of the simulation performed with respect to each of the proposal method based on a Range-Based system and a Range-Free system is shown. It was confirmed that position estimation was possible with a particle filter, and that the accuracy of estimation was improved by the method used in Active learning rather than randomly selecting an observation location. As a problem setting of the simulation, the number of particles of the particle filter is 1,000, and the initial values of the parameters and the position of the signal source (x _t , y _t ) are randomly selected within the range of Equation 12.
Further, the signal source constants are set to P ₀ = 60.0 and κ = 2.2. For simplicity, beacons and receivers are assumed to be omnidirectional. For observation, Gaussian white noise with an average of 0 and a variance of 1 is placed. It is assumed that the receiver cannot observe a signal with a signal strength of less than 0 dB, and treats all as P = −30 in the calculation.

The simulation results of the proposed method based on the Range-Based method are shown in Table 1 and Table 2 (Tab1, Tab2) below. Δ in the table is an estimation error.
Here, the simulation is performed 100 times for each number of observations, and the average value is listed. It can be confirmed that high-precision estimation can be realized with a small number of observations by selecting according to uncertainty sampling rather than randomly selecting an observation location. Further, FIG. 3 shows a state in which the next observation location can be appropriately selected by using uncertainity sampling. In the figure, the red star is the target position, the blue circle is the sensor position, and the green circle is the particle position. FIG. (A) is a state at a certain time t = T, and the state after the observation place is randomly selected from the state at the next time and observation / resampling is performed is FIG. (B). FIG. 5C shows the state after the observation location is selected by uncertain sampling and the observation / resampling is performed.

  As described above, in the present invention, it has been shown that the estimation can be performed online by using the Particle Filter when estimating the position of the signal source. Furthermore, when determining the observation location, it was confirmed that it can be estimated efficiently with few observations by deciding according to Active learning, and its usefulness was also shown. However, when estimating the position of a victim, the most important thing is not to reduce the number of observations, but to shorten the estimation time. Reducing the number of observations is important for shortening the estimation time, but does not necessarily lead to a reduction in the estimation time. This is because the observation location itself is small, but it takes time to move due to factors such as the observation locations being very far apart. Therefore, it is necessary to build a model that takes into account the travel time between observation sites as a future issue.

  The unmanned moving object of the present embodiment is a thing that flies, but any thing that can be moved independently, such as a thing that moves on the ground or a thing that moves underwater, can be used. Applicable.

  The above-described embodiments are merely examples for facilitating understanding of the present invention, and are not intended to limit the present invention. The present invention can be changed and improved without departing from the gist thereof, and it is needless to say that the present invention includes equivalents thereof.

1 Unmanned flying vehicle 2 Management server

Claims (2)

A method for estimating the position of an object by an unmanned moving object,
A step of tentatively determining an estimated position of the object; a step of observing a signal transmitted from the object by the unmanned mobile body; a step of updating the estimated position and determining a next observation place after the update; Moving to a later observation location and performing the observation again, and gradually estimating the position of the object,
Position estimation method. The position estimation method according to claim 1,
The step of observing the signal uses a particle filter, and the step of determining the next observation location uses active learning.
Position estimation method.