JP5396311B2 - Behavior prediction apparatus, method, and program - Google Patents

Description

  The present invention relates to a behavior prediction apparatus, method, and program for estimating a route to reach a destination and a time at which each user passes along a route along the route.

  There is a navigation service that allows the user to obtain the route to the destination, the location where the user passes along the route, and the time of arrival at the destination by inputting the user's current position and destination. As a technique for realizing such navigation, Patent Document 1 and Patent Document 2 describe a method for accurately predicting a travel time to a destination on the assumption that a route through which a user passes is known.

Japanese Patent Laid-Open No. 11-328572 JP 2002-117492 A

  Existing navigation systems assume that users act along a certain route. However, in this method, when there are a plurality of routes arriving at the destination, or when the user does not act according to the route presented by the navigation system, estimation of the time and route arriving at the destination fails.

  The present invention has been made in view of the above circumstances, and provides a behavior prediction apparatus, method, and program for estimating a route to reach a destination and a time at which each user passes along the route. The purpose is to do.

  In order to solve the above-described problem, the behavior prediction apparatus according to the present invention predicts the start time and the position and time of the user when the user moves from the start position at the time to the destination position of the destination. A plurality of particles including a first time, a first position at the first time, a first velocity at the first time and the first position, and a first weight score; Setting all weight scores of the plurality of particles at the start time to zero, and setting all positions, velocities, and times of the plurality of particles to the start position, the speed at the start position, and the start time, respectively. And a distribution function for each particle, calculating a second position of the particle at a second time in the future based on the current position and current velocity of the particle, and averaging the current velocity of the particle Based on the second speed of the particle at the second time, a past location time of the user among the plurality of particles at the second time, a location at the location time, the location time, and the A calculation means for calculating a second weight score having a higher value for a particle having a value closer to the past position information history including the location speed at the location, and a second location where the distribution density of the plurality of particles is the highest. Output means for outputting the position estimation result of the user at the time.

  According to the present invention, it is possible to estimate a route to reach the destination and a time at which the user passes through each point along the route.

The block diagram of the action prediction apparatus which concerns on embodiment of this invention. The flowchart which shows an example of operation | movement of the action prediction apparatus of FIG. The flowchart which shows a detailed example of step S204 of FIG.

Hereinafter, an action prediction apparatus, method, and program according to embodiments of the present invention will be described in detail with reference to the drawings. Note that, in the following embodiments, the same numbered portions are assumed to perform the same operation, and repeated description is omitted.
First, an outline of the operation principle of the behavior prediction apparatus according to the embodiment will be described.
By using the position information history acquired by the position information measurement system mounted on the device carried by the user, the behavior prediction apparatus according to the embodiment uses the user's route based on the history when the user has moved in the past. In addition, it is possible to estimate the time when the vehicle passes the moving waypoint. In the present embodiment, a particle filter is used for estimating a route on which the user behaves. By using the particle filter, it is possible to automatically select a history useful for estimating the action path from the past action paths of the user, and to predict the path. In addition, by using the particle filter, it is possible to probabilistically determine which route is taken when there are several types of routes taken by the user.

Next, the behavior prediction apparatus according to the embodiment will be described with reference to FIG.
The behavior prediction apparatus according to the present embodiment includes a position query input unit 101, a position information history recording unit 102, a behavior prediction processing unit 103, and a prediction result output unit 104.
The position query input unit 101 is a set of coordinates of the user's current stay position (specified by latitude and longitude) and the current time, and the coordinates of the user's destination (latitude and longitude), which trigger the behavior prediction process. Specified) as input. A set of the coordinates of the current stay position and the current time is represented by (latitude, longitude, measured time), and the coordinates of the user's destination are represented by (latitude, longitude). The behavior prediction process is performed by passing the received query to the behavior prediction processing unit 103.

  The location information history recording unit 102 records user location information (also referred to as location information history) recorded in the past. The recorded position information consists of a triple of latitude, longitude and measured time, and is represented by (latitude, longitude, measured time). Further, it is assumed that the position information is continuously recorded for a long period of time, for example, one month or more by a position information measurement system mounted on a device carried by the user.

  The behavior prediction processing unit 103 performs behavior prediction processing using the input received from the location query input unit 101 and the user location information history stored in the location information history recording unit 102. The prediction result of the user's position at a certain time is expressed as a set of points having latitude and longitude values called a finite number of particles. A place with a higher particle distribution density is interpreted as a place where the user is more likely to be at that time. The prediction result is passed to the prediction result output unit 104. Details of the particles will be described later with reference to FIG.

  The prediction result output unit 104 estimates and outputs the position of the user from the set of particles received from the behavior prediction processing unit 103. For the estimation of the user position, for example, a method of taking the particle centroid or a method of calculating a probability distribution of the user position by a kernel density estimation method can be applied.

Next, an example of operation | movement of the action prediction apparatus of embodiment is demonstrated with reference to FIG.
After the process starts, the position query input unit 101 receives the input query (step S201). The input query includes the coordinates of the current position of the user and the time s = (x _s, y _s, t _s ), and the coordinates of the target staying place g = (x _g, y _g ).

The behavior prediction processing unit 103 performs the particle filter initialization process based on the parameters received as input in step S201 (step S202). Here, p _t ⁽ⁱ⁾ is the i-th particle at time t, and p _t ⁽ⁱ⁾ = <x ⁽ⁱ⁾ _, y ⁽ⁱ⁾ _, u ⁽ⁱ⁾ _, v ⁽ⁱ⁾ _, w ⁽ⁱ⁾ _, It is defined as t>. Here, x ⁽ⁱ⁾ and y ⁽ⁱ⁾ are the longitude and latitude of the particles p _t ⁽ⁱ⁾ , u ⁽ⁱ⁾ and v ⁽ⁱ⁾ are the east-west and north-south velocities, respectively, and w ⁽ⁱ⁾ is the weight. The score, t is time. Assuming that the number of particles to be generated is N, the particle initialization is to set the weight score to zero for each of p _ts ⁽¹⁾ ,..., P _ts ^(N) , that is, p _ts ⁽ⁱ⁾ = <x _{s ,} Y _s, u _s, v _s, 0 _, t _s > (i is a natural number from 1 to N). Here, the vector (u _s, v _s ) is a unit vector that is a multiple of the direction vector connecting s and g. In the following, vector x ⁽ⁱ⁾ = (x ⁽ⁱ⁾ _, y ⁽ⁱ⁾ ), vector v ⁽ⁱ⁾ = (u ⁽ⁱ⁾ _, v ⁽ⁱ⁾ ), vector v _s = (u _s, v _s ) Is also used in combination with a notation using a vector. In the figures and mathematical formulas, vectors are simply shown in bold. For example, the vector x is simply indicated by bold x.

  The behavior prediction processing unit 103 determines whether the set of particles at a certain time t has reached the destination g or whether the number of iterations from step S204 to step S206 is equal to or greater than a predetermined threshold value. (Step S203). Here, whether or not the destination g has been reached can be determined by, for example, determining whether or not the ratio of the particles within the distance ε and the particles outside the distance ε from g exceeds a certain threshold value. The behavior prediction processing unit 103 determines that the destination g has been reached when this ratio is equal to or greater than the threshold value. If the determination result is true, the process ends. If the determination result is false, the process proceeds to step S204.

The behavior prediction processing unit 103 performs a particle update process (step S204).
Here, the details of the particle update processing in step S204 will be described with reference to the flowchart of FIG.
First, a new position of the particle at time t + 1 is calculated based on the current position and speed (step S301). Here, Δt is a time width between time t and time t + 1, and is given as a parameter at the start of processing.

  Next, the speed vector is updated (step S302). Here, N (μ, Σ) represents a two-dimensional Gaussian distribution having an average value μ and a variance-covariance matrix Σ, and this expression generates a normal random number along the two-dimensional Gaussian distribution, and The value is the updated value of the velocity vector. The variance-covariance matrix is given as a parameter at the start of processing.

Finally, the particles are updated with the speed and position updated in step S302 (step S303), and the process ends.
Returning again to the flowchart of FIG. 2, the process proceeds to step S205.
The behavior prediction processing unit 103 calculates a weight score w ′ ⁽ⁱ⁾ for each particle (step S205). For the calculation of the weight score, position information history data acquired in the past is used. Here, it is assumed that the vector x included in the position information history, which is past observation data, is associated with the latitude, longitude, and time as well as the speed at that time. The speed may be measured at the same time when the position is acquired, or the behavior prediction apparatus according to the embodiment may calculate the difference from the position and time difference with the observation data immediately before and after.

A set of all past observation data is D, and each element d of D (dεD) is d = <vector x _d, vector v _d, t _d >, and a set of position, velocity, and observed time To do. The calculation formula of the weight score using D is as the following formula (1).

ここで、ｓｉｍはコサイン類似度を［０_，１］の区間に移した関数であり、下記の（２）式で表される。

Here, sim is a function in which the cosine similarity is moved to the interval [0 _, 1], and is expressed by the following equation (2).

ここでθ_１、θ_２は重みスコア計算におけるパラメータであり、本手法実行時に手動で定めるものとする。
次に行動予測処理部１０３が、現在のパーティクルの集合｛ｐ_ｔ ^（１） _，ｐ_ｔ ^（２） _，…_，ｐ_ｔ ^（Ｎ）｝から、１つのパーティクルが、それぞれ確率ｗ^（１）／Ｗ_ｔ、ｗ^（２）／Ｗ_ｔ、…、ｗ^（Ｎ）／Ｗ_ｔで選択されるとしてＮ回復元抽出を行い、新たなＮ個のパーティクルの集合を生成し獲得する（ステップＳ２０６）。ここでＷ_ｔ＝Σｗ^（ｉ）である。
ステップＳ２０４からＳ２０６までを繰り返す回数が時間を進めることになるので、この回数を調整すれば任意の時点でのユーザの位置の予測が可能になる。

Here, θ ₁ and θ ₂ are parameters in the weight score calculation, and are manually determined when this method is executed.
Next, the behavior prediction processing unit 103 determines that each particle has a probability w ⁽¹⁾ / W from the current set of particles { _pt ⁽¹⁾ _{, pt} ⁽²⁾ _, ... _{, pt} ^(N) }. Extraction is performed N times assuming that _t , w ⁽²⁾ / W _t ,..., w ^(N) / W _t is selected, and a new set of N particles is generated and acquired (step S206). Here, W _t = Σw ⁽ⁱ⁾ .
Since the number of times of repeating steps S204 to S206 advances the time, adjusting the number of times makes it possible to predict the position of the user at an arbitrary time.

Further, the method of the present embodiment can also be applied to an interpolation problem that estimates a user's position between two times when the user's position is observed at a certain two times. In that case, the behavior predicting device of FIG. 1 may be executed with the earlier one of the two points to be interpolated as the start point and the latter as the destination point. Furthermore, for the particle distribution obtained by the processing, “Mike Klaas, Mark Briers, Nando de Freitas, Arnaud Doucet, Simon Maskell, Dustin Lang,“ Fast Particle Filter Smoothing: If I Had a Million Particles ”, In proceedings of by applying ^{the 23 rd international conference on machine learning} , Backward Smoother , such as in 2006 ", it is possible to re-calculation starting point, the distribution of particles incorporating the constraints the coordinates of the destination point.

  According to the behavior prediction apparatus described above, the user's position is predicted at an arbitrary time in the future using the accumulated past user position information history, and at an arbitrary time between two observation points. Position prediction can be performed. According to the behavior predicting device of the embodiment, for example, when the user moves using public transportation means, when the current position of the user and the destination to be reached are known, the route to reach the destination , And the time at which the user passes through each point along the route can be estimated.

The instructions shown in the processing procedure shown in the above embodiment can be executed based on a program that is software. The general-purpose computer system stores this program in advance and reads this program, so that it is possible to obtain the same effect as that obtained by the behavior prediction apparatus of the above-described embodiment. The instructions described in the above-described embodiments are, as programs that can be executed by a computer, magnetic disks (flexible disks, hard disks, etc.), optical disks (CD-ROM, CD-R, CD-RW, DVD-ROM, DVD). ± R, DVD ± RW, etc.), semiconductor memory, or a similar recording medium. As long as the recording medium is readable by the computer or the embedded system, the storage format may be any form. If the computer reads the program from the recording medium and causes the CPU to execute instructions described in the program based on the program, the same operation as that of the behavior prediction apparatus of the above-described embodiment can be realized. Of course, when the computer acquires or reads the program, it may be acquired or read through a network.
In addition, the OS (operating system), database management software, MW (middleware) such as a network, etc. running on the computer based on the instructions of the program installed in the computer or embedded system from the recording medium implement this embodiment. A part of each process for performing may be executed.
Furthermore, the recording medium in the present invention is not limited to a medium independent of a computer or an embedded system, but also includes a recording medium in which a program transmitted via a LAN or the Internet is downloaded and stored or temporarily stored.
Further, the number of recording media is not limited to one, and when the processing in the present embodiment is executed from a plurality of media, it is included in the recording media in the present invention, and the configuration of the media may be any configuration.

The computer or the embedded system in the present invention is for executing each process in the present embodiment based on a program stored in a recording medium, and includes a single device such as a personal computer or a microcomputer, Any configuration such as a system in which apparatuses are connected to a network may be used.
Further, the computer in the embodiment of the present invention is not limited to a personal computer, but includes an arithmetic processing device, a microcomputer, and the like included in an information processing device, and a device capable of realizing the functions in the embodiment of the present invention by a program, The device is a general term.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

DESCRIPTION OF SYMBOLS 101 ... Position query input part, 102 ... Position information history recording part, 103 ... Action prediction process part, 104 ... Prediction result output part.

Claims (7)

A behavior prediction device that estimates a start time and a user's position and time when the user moves from the start position at the time to the destination position of the destination,
A plurality of particles including a first time, a first position at the first time, a first speed at the first time and the first position, and a first weight score are set, and at the start time Setting means for setting all the weight scores of the plurality of particles to zero, and setting all positions, velocities, and times of the plurality of particles to the start position, the speed at the start position, and the start time, respectively;
For each particle, calculate the second position of the particle at the second future time based on the current position and current velocity of the particle, and at the second time based on a distribution function that averages the current speed of the particle. The second speed of the particle is calculated, and among the plurality of particles at the second time, the past location time for the user, the location at the location time, the location time and the location speed at the location position Calculating means for calculating a second weight score having a higher value for a particle having a value closer to a past position information history including:
And an output means for outputting a place where the distribution density of the plurality of particles is highest as a user position estimation result at the second time. The calculation means uses the following w ′ ⁽ⁱ⁾ as the second weight score.

I is a particle index, and D is a set of data included in all past position information histories, and each element d (dεD) of D is d = <vector x _d , vector v _{where d 1} , t _d >, t _d is the past location time, vector x _d is the location at the location time, and vector v _d is the location time and the location speed, and the vector v _s is It is the speed at the start time, θ ₁ and θ ₂ are parameters having certain values, and sim is a function that moves the cosine similarity to the interval of [0 _, 1],

The behavior prediction apparatus according to claim 1, wherein When the positions of the plurality of particles at the second time can be regarded as being within the threshold distance from the target position, the second time is determined as the target arrival time, and can be regarded as being within the threshold distance. If not, the second time is set as the current time, the second position and the second speed calculated by the calculation means are set as the current position and the current speed, respectively. Determination means for calculating three positions, a third speed, and a third weight score;
The probability calculated for each particle from all particle sets having the third position, the third speed, and the third weight score (corresponding third weight score) / (the third weight score of all particles) 3. The behavior according to claim 1, further comprising: generating means for performing restoration extraction by the number of particles and generating a new particle set assuming that the particles are selected in (sum). Prediction device. A behavior prediction method that estimates a start time and a user's position and time when the user moves from the start position at the time to the destination position of the destination,
A plurality of particles including a first time, a first position at the first time, a first speed at the first time and the first position, and a first weight score are set, and at the start time A setting step for setting all weight scores of the plurality of particles to zero, and setting all positions, velocities, and times of the plurality of particles to a start position, a speed at the start position, and a start time, respectively;
For each particle, calculate the second position of the particle at the second future time based on the current position and current velocity of the particle, and at the second time based on a distribution function that averages the current speed of the particle. The second speed of the particle is calculated, and among the plurality of particles at the second time, the past location time for the user, the location at the location time, the location time and the location speed at the location position A calculation step of calculating a second weight score having a higher value for a particle having a value closer to a past position information history including:
An output step of outputting a place having the highest distribution density of a plurality of particles as a position estimation result of the user at the second time. In the calculation step, as the second weight score, the following w ′ ⁽ⁱ⁾

I is a particle index, and D is a set of data included in all past position information histories, and each element d (dεD) of D is d = <vector x _d , vector v _{where d 1} , t _d >, t _d is the past location time, vector x _d is the location at the location time, and vector v _d is the location time and the location speed, and the vector v _s is It is the speed at the start time, θ ₁ and θ ₂ are parameters having certain values, and sim is a function that moves the cosine similarity to the interval of [0 _, 1],

The behavior prediction method according to claim 4 , wherein: When the positions of the plurality of particles at the second time can be regarded as being within the threshold distance from the target position, the second time is determined as the target arrival time, and can be regarded as being within the threshold distance. If not, the second time and the second speed calculated in the calculation step are set as the current time and the current speed as the current position and the current speed, respectively. A determination step for calculating three positions, a third speed, and a third weight score;
The probability calculated for each particle from all particle sets having the third position, the third speed, and the third weight score (corresponding third weight score) / (the third weight score of all particles) performed only re-sampling the number of particles as the particles in the sum) is selected, the behavior of claim 4 or claim 5, characterized in that further comprising a generation step of generating a new particle aggregate, the Prediction method.   The program for making a computer run as each means of the action prediction apparatus of any one of Claims 1-3.

Images