JP4627256B2 - Movement prediction apparatus, movement prediction method, and program - Google Patents

Description

  The present invention relates to a movement prediction technique, and more particularly, to a movement prediction technique for predicting the positions of moving humans, vehicles, and objects (objects).

  In recent years, with the spread of devices that measure positions, it has become possible to track various moving objects such as people, vehicles, and objects that move in the real world. Development of MODB (Moving Object DataBase) that accumulates data of such moving objects is in progress (see, for example, Non-Patent Document 1).

One of the research subjects of such position information database is how to efficiently create an index (see, for example, Non-Patent Documents 2-4).
The moving object data is a time series of combinations of time t and coordinate values x, y such as (t, x, y). Hereinafter, this data is referred to as trajectory data. Since the trajectory data is a kind of spatial data, it is possible to apply a spatial index method such as R-Tree using MBR (Minimum Bounding Rectangle) (for example, see Non-patent Document 5). . However, since the trajectory data is continuously added as time passes, there is a problem that the frequency of recalculating MBR becomes high in the normal R-Tree.

  Conventionally, as a method for solving this problem, STP-Tree (see, for example, Non-Patent Document 6), TPR-Tree (for example, refer to Non-Patent Document 6) as a method for creating an MBR so as to include a future position by predicting a destination of an object. For example, methods such as Non-Patent Document 7) and TPR * -Tree (for example, refer to Non-Patent Document 1) have been proposed. If an MBR is created so as to include data that will be added in the future, it is not necessary to rebuild the index even if new data is added over time. That is, in order to reduce the index reconstruction frequency by this method, it is important to accurately predict the position of a future object.

Next, with reference to FIG. 7, the position prediction method by TRP-Tree is demonstrated. FIG. 7 is an explanatory diagram showing a position prediction method based on TRP-Tree.
In TPR-Tree and TPR * -Tree, the maximum speed and the minimum speed from a certain time t to the past t-m are obtained for each object for each x-axis and y-axis. These are assumed to be V _{x +} , V _{y +} , V _x− , V _y− . In TPR-Tree, the quadruple of V _{x +} , V _{y +} , V _x− and V _y− is called VBV (Velocity Bounding Vector).

At this time, since only the maximum speed is used for prediction, the accuracy of predicting the position of the single object is low. Therefore, when used for actual index creation, the positive and negative of the x-axis and y-axis are used for a plurality of objects included in the MBR. Find the maximum speed for each direction and negative direction. Similarly, V _x− , V _{y +} and V _y− are also obtained.

If the object does not advance in the negative direction of the x axis, V _x− is the minimum velocity when the object advances in the positive direction, and the same applies to the y axis. For this reason, in the example of FIG. 7, V _{y +} is a negative value.
Using these values, x and y coordinates P _{x +} (t + i), P _{y +} (t + i), P _x− (t + i), P _y of the four corner points of the MBR at time t + i (where t and i are positive integers) _- (T + i) is obtained as follows. P _x (t) and P _y (t) indicate the coordinates of the object at time t.
P _{x +} (t + i) = V _{x +} × i + P _x (t)
P _{y +} (t + i) = V _{y +} × i + P _y (t)
P _x− (t + i) = V _x− × i + P _x (t)
P _y− (t + i) = V _y− × i + P _y (t)

Next, a position prediction method using STP-Tree will be described.
In STR-Tree, the position from a certain time t to the past t−m (where m is a positive integer) is set to x (t−m), y (t−m), x (t−m + 1), and y ( t−m + 1),..., x (t), y (t). A vector extracted from time t-m to t is denoted as k (t) _m, and a vector extracted from time t−m−1 to t−1 is denoted as k (t−1) _m . The number of elements is always 2m. A 2m × n matrix with k (t) _m as the top row and the rows from k (t-1) _m to k (tn) _m (n <m) (where n is a positive integer) arranged in order. Is k (t) _{m, n} . Separately from this, let x (t) _n = {x (t−n),..., X (t)} be a vector in which the values of x coordinates from time t to t−n are arranged. The same applies to y (t) _n .

At this time, an approximate solution vector w _x = {wx ₁ , wx ₂ ,..., Wx _2m } satisfying the following simultaneous equations, w _y = {wy ₁ , wy ₂ ,. _2m }.
x (t) ^T = K (t-1) _{m, n} · w _x ^T
y (t) ^T = K (t-1) _{m, n} · w _y ^T

The obtained matrix w _x, w _y and a vector _{k m (t) = {x} (t-m), y (t-m), x (t-m + 1), y (t-m + 1), .. .., x (t), y (t)}, the position p (t + 1) = {x (t + 1), y (t + 1)} of the object at time t + 1 is obtained by the following equation.
x (t + 1) = w _x · k (t) _m ^T
y (t + 1) = w _y · k (t) _m ^T
The position p (t + 2),... After the time t + 2 can also be obtained by applying this expression recursively.

Ouri Wolfson, Prasad Sistla, Bo Xu, Jutai Zhou, and Sam Chamberlain.DOMINO: Databases for Moving Objects tracking.In SIGMOD'99 Conference Proceedings, pp.547-549, 1999. Hoda Mokhtar, Jianwen Su, and Oscar H. Ibarra.On moving object queries.In POPS2002 Symposium Proceedings, PP.188-198, 2002. George Kollios, Dimitrios Gunopulos, and Vassilis J. Tsotras.On indexing mobile oblects.In PODS'99 Symposium Proceedings, pp.261-272, 1999. G.Kollios, VJTsotras, D.Gunopulos, A.Delis, and M.Hadjieleftheriou.Indexing animated objects using spatiotemporal access methods.IEEE Transactions on Knowledge and Data Engineering, Vol.13, No.5, pp.758-777, 2001. O. Guttman.R-trees: a dynamic index structure for spatial searching.In SIGMOD'84 Conference Proceedings, pp.47-57, 1984. Yu Fei Tao, Christos Faloutsos, Dimitris Papadias, and Bin Liu.Prediction and indexing of moving objects with unknown motion pattens.In SINGMOD'04: Proceedings of then 2004 ACM SIGMOD international conference om Management of data, pp.611-622, New York , NY, USA, 2004. ACM Press. Simonas Saltenis, Christian S. Jensen, Scott T. Leutenegger, and Mario A. Lopez.Indexing the positions of continuously moving objects. , New York, NY, USA, 2000. ACM Press. Yufei Tao, Jimeng Sun, and Dimitris Papadias.Analysis of predictive spatio-temporal queries.ACM Trans.Database Syst., Vol28, No.4, pp.295-336, 2003.

However, such a conventional technique has a problem that a good accuracy rate cannot be obtained for the movement prediction of a moving object.
For example, TPR-Tree shows a relatively good predicted accuracy rate when the object is moving straight ahead or completely stopped. Even in the case where the moving direction is indefinite, such as a random walk, the prediction accuracy rate does not drop so much because the prediction range is large as a result. On the other hand, because the influence of position fluctuation (measurement error) is not taken into account, even if it is in a stagnation state, if the position fluctuation is large, the prediction accuracy rate decreases or an MBR with an excessive area is created. There was a problem that it often ends up.

  In SPR-Tree, the predicted object position is given by afin transformation. For this reason, in addition to stopping and going straight, a high predicted accuracy rate is shown for an object that moves like a circular motion or a sine curve. On the other hand, if there is an inversion or a turn, the approximate solution may not be an appropriate value, and the predicted correct answer rate may drop significantly. In addition, since the influence of the position fluctuation is not taken into consideration as in the case of TPR-Tree, if there is a stagnation including fluctuation, the prediction accuracy rate decreases. Further, in this method, it is necessary to perform singular value decomposition in order to obtain a solution vector (solve simultaneous equations). However, the algorithm is generally complicated and has a problem of a large amount of calculation.

  The present invention is for solving such a problem. Even when the movement destination of an object is predicted using data including position fluctuation (measurement error), high prediction accuracy can be obtained with a relatively small amount of calculation. It is an object to provide a movement prediction apparatus, a movement prediction method, and a program.

In order to achieve such an object, a movement prediction apparatus according to the present invention includes an arithmetic processing unit that performs arbitrary arithmetic processing on input data, and a storage unit that stores data used in the arithmetic processing unit. A movement prediction device that calculates a movement prediction range of the moving object based on the locus data including the position information of the moving object at each time by the arithmetic processing unit , and the trajectory data of the moving object is externally input to the arithmetic processing unit. Data acquisition means for acquiring and storing in the storage unit, and the position vector p (t−m) of the moving object from time t−m (where t and m are positive integers) included in the trajectory data of the storage unit from time t To the moving velocity vector v (i) = p (i) −p (i−1) at each time i (where t−m ≦ i ≦ t: i is a positive integer) based on p (t) Memory A moving speed calculating means for storing the stagnation locus data of the moving object in the serial憶部, straight, and moves to identify the moving state of the moving object by determining it meets any of the conditions indicating each moving state of Ranpo a state specifying means, in accordance with the moving state of a specific has been moving object and a moving prediction range calculation means for calculating a movement prediction range of the moving object.

In addition, the moving state specifying means includes a moving object position vector p (t−m) to p (t) from time t−m (where t and m are positive integers) included in the trajectory data. And the position fluctuation width θ _p of the moving object having a larger value of the measurement error of the sensor that measured the trajectory data or the diameter of the moving object, | p (i) −p (t) | <θ _p If the condition of t−m ≦ i ≦ t: i is a positive integer is satisfied, it is determined that the moving object is in a stagnation state. j (provided that t <j ≦ t + n: j, n is a positive integer) predicted moving range of the moving object in is obtained by such a region within a circle of radius theta _p centered at least p (t) of There is .

In addition to this, the moving state specifying means determines that the moving speed vector v (t−m) to v (t) and the position fluctuation width θ _p of the moving object are | v (i) −v (t ) | <Θ _p (where t−m ≦ i ≦ t: i is a positive integer), the moving object is determined to be in a straight traveling state, and the moving predicted range calculation means determines that the moving object is in a straight traveling state. , The movement prediction range of the moving object at time j (where t <j ≦ t + n: j and n are positive integers) is centered on at least p (t) + (j−t) v (t). it is obtained as the area within a circle of radius theta _p.

In addition to this, the moving state specifying means uses the moving speed vectors v (t−m) to v (t) of the storage unit, the maximum moving speed v _max of these moving speed vectors, and the position fluctuation width θ _p of the moving object. However, if the condition | v (i) | <v _max + θ _p (where t−m ≦ i ≦ t: i is a positive integer) is satisfied, it is determined that the moving object is in a random walking state, and the movement prediction range is calculated. When the moving object is in a random walking state, the movement predicted range of the moving object at time j (where t <j ≦ t + n: j, n is a positive integer) is at least a radius centered on p (t) ( it is obtained as a _{j-t) v max + θ} p region within a circle.

At this time, the movement prediction range calculation means may set the probability of existing at the position of radius r from p (t) at time j to (r / (j−t) v _max ) (j−t). Good.

In addition, the movement prediction method according to the present invention includes a calculation unit that performs arbitrary calculation processing on input data, and a movement prediction device that includes a storage unit that stores data used in the calculation processing unit. a movement prediction method for calculating the predicted moving range of the moving object based on the locus data including the position information at the time, the data acquisition processing unit, which obtains and stores locus data of the moving object from the outside to the storage unit The step and the calculation processing unit include the position vector p (tm) to p (t) of the moving object from time t-m (where t and m are positive integers) included in the trajectory data of the storage unit. Based on the above, a moving velocity vector v (i) = p (i) −p (i−1) at each time i (where t−m ≦ i ≦ t: i is a positive integer) is calculated and stored in the storage unit. Particular the moving speed calculating step, the arithmetic processing unit, stagnation of the locus data in the storage unit is the moving object, straight, and the moving state of the moving object by determining it meets any of the conditions indicating each moving state of the random walk that a movement status determining step of the arithmetic processing unit, in accordance with the moving state of the identified moving object and a moving prediction range calculation step of calculating a movement prediction range of the moving object, in a movement status determining step, locus data , And the measurement error or movement of the sensor that measured the trajectory data from the time t-m (where t and m are positive integers) to the position vector p (t-m) to p (t) of the moving object from time t the width theta _p position fluctuation of the moving object consisting of any larger value of the diameter of the object, | p (i) -p ( t | <Θ _p (except t-m ≦ i ≦ t: i is a positive integer) If the condition is satisfied, the said moving object is determined to be in the stagnant state, the moving velocity vector v of the storage unit (t-m) v (t) and the position fluctuation width θ _p of the moving object satisfy the condition of | v (i) −v (t) | <θ _p (where t−m ≦ i ≦ t: i is a positive integer). In this case, it is determined that the moving object is in a straight traveling state, the moving speed vectors v (t−m) to v (t) of the storage unit, the maximum moving speed v _max of these moving speed vectors , and the position fluctuation of the moving object When the width θ _p satisfies the condition of | v (i) | <v _max + θ _p (where t−m ≦ i ≦ t: i is a positive integer), the moving object is determined to be in a random walking state, and moved. In the predicted range calculation step, the moving state identification step stops the moving object. If it is determined that the state, the time j (except t <j ≦ t + n: j, n is a positive integer) predicted moving range of the moving object in the inside circle of radius theta _p centered at least p (t) of When the moving state specifying step determines that the moving object is in the straight traveling state, the movement prediction range of the moving object at time j (where t <j ≦ t + n: j and n are positive integers) is at least p. (t) + (j-t ) v (t) is an area within a circle having a radius theta _p centered, if the movement status determining step determines that the moving object is in the random walk state, the time j (except t < The movement prediction range of the moving object when j ≦ t + n: j and n are positive integers) is set to be an area within a circle having a radius (j−t) v _max + θ _p centered at least on p (t) . Is.

Moreover, the program concerning this invention is a program for functioning as each part which comprises one of the movement prediction apparatuses mentioned above.

  According to the present invention, the movement prediction device identifies the moving state of the moving object by determining which of the conditions indicating the moving state of the moving object is stagnation, straight ahead, and random walking by the movement prediction device. Since the movement prediction range of the moving object is calculated according to the state, the movement prediction range can be calculated considering the position fluctuation appropriately according to the movement state of the moving object, and data including position fluctuation (measurement error) is obtained. Even when the movement destination of an object is predicted by using it, high prediction accuracy can be obtained with a relatively small amount of calculation.

Next, embodiments of the present invention will be described with reference to the drawings.
[First Embodiment]
First, with reference to FIG. 1, a moving object database system in which the movement prediction apparatus according to the first embodiment of the present invention is used will be described. FIG. 1 is a block diagram showing a configuration of a moving object database system in which the movement prediction apparatus according to the first embodiment of the present invention is used.

A moving object database system (hereinafter referred to as a moving object DB system) accumulates trajectory data related to an input arbitrary moving object, searches for a moving object having a similar trajectory in response to a search request from the outside, and performs the search. It is a system that outputs results.
This moving object DB system includes a trajectory data input device 1, an object data management device 2, a movement prediction device 3, and a data search device 4.

The trajectory data input device 1 includes a data acquisition unit 11 as a main functional unit.
The data acquisition unit 11 has a function of receiving the trajectory data 5 of the moving object measured by the sensor and a function of collectively outputting the position data 1A indicating the position of each object at each time to the object data management device 2. ing. The trajectory data 5 can be received via a communication network or acquired as an electronic file from an external device or a recording medium.

The object data management device 2 includes a data generation unit 21, a data search unit 22, and a data storage unit 23 as main functional units.
The data generation unit 21 has a function of creating an index for the position data 1A received from the trajectory data input device 1, a function of saving the position data 1A and object data including the index in the object data storage unit 23, and an index creation time. When it is necessary to predict movement of an arbitrary object, a prediction request 2A including past position data necessary for prediction is output to the movement prediction device 3, and based on a prediction region list included in the obtained prediction result 2B. And a function for creating an index.

The data search unit 22 includes a function for searching desired object data by referring to the index of each object from the object data storage unit 23 in response to a search request 4A from the data search device 4, and the obtained object data. A function of returning the search result 4B to the data search device 4;
The data storage unit 23 uses a storage device such as a hard disk to store the object data generated by the data generation unit 21 so as to be searchable by its index, and any object according to a request from the data search unit 22 A function of reading and outputting data.

The movement prediction device 3 includes a movement prediction unit 31 as a main functional unit.
The movement prediction unit 31 uses a function for specifying a movement state of an arbitrary object based on a prediction request 2A from the object data management device 2 and a prediction area indicating a movement destination of the object using a prediction function corresponding to the movement state. And a function of returning a prediction area list including the obtained prediction areas to the object data management apparatus 2 as a prediction result 2B.

The data search device 4 includes a search control unit 41 as a main functional unit.
The search control unit 41 has a function of outputting a search request 4A for instructing search of object data corresponding to the time, position or range, and object name specified in the search request 6B to the object data management device 2, and an object data management device 2 has a function of displaying the search result received from 2 on the screen, outputting it as electronic data to an external device or a recording medium, or transmitting it to a communication network.

  The trajectory data input device 1, the object data management device 2, the movement prediction device 3, and the data search device 4 constituting the moving object DB system according to the present embodiment are realized by an information processing device such as a computer. Each functional unit includes a microprocessor such as a CPU and an arithmetic processing unit having peripheral circuits thereof, which read and execute a program stored in the storage unit of the device, whereby the hardware and the program cooperate with each other. Realized. In the present embodiment, the case where these devices are realized by separate devices has been described as an example. However, the present invention is not limited to this, and a plurality of devices may be realized by one device.

[Operation of First Embodiment]
Next, the operation of the moving object DB system according to the first embodiment of the present invention will be described with reference to FIG.
The main operation of the moving object DB system includes an object data storing operation for storing object data related to each moving object and an object data searching operation for searching arbitrary object data.

[Object data storage operation]
First, an object data storing operation for storing object data related to each moving object in the moving object DB system will be described.
The trajectory data input device 1 sequentially receives the trajectory data 5 of each moving object measured by the sensor by the data acquisition unit 11, and collects the position data 1A indicating the position of each object at each time to the object data management device 2. Output.

The object data management device 2 uses the data generation unit 21 to create an index for the position data 1A received from the trajectory data input device 1, and stores the position data 1A and object data including the index in the object data storage unit 23.
At this time, the data generation unit 21 outputs the prediction request 2A including the past position data necessary for the prediction to the movement prediction device 3 when it is necessary to predict the movement of an arbitrary object at the time of index creation. An index is created based on the prediction area list included in the prediction result 2B.

In response to the prediction request 2A from the object data management device 2, the movement prediction device 3 uses the movement prediction unit 31 to execute a movement prediction process in FIG. 2 described later to identify the movement state of an arbitrary object. A prediction area indicating the movement destination of the object is calculated using a prediction function corresponding to the movement state. Then, the prediction area list including the obtained prediction areas is returned to the object data management apparatus 2 as the prediction result 2B.
Thus, the position and index of each moving object are generated based on the trajectory data input at each time, and the object data of each moving object stored in the data storage unit 23 is updated.

[Object data search operation]
Next, an object data search operation for searching arbitrary object data from the moving object DB system will be described.
The data search device 4 searches for object data corresponding to the time, position or range, and object name specified by the search request 6B by the search control unit 41 in response to a search request 6A input from an operator operation or an external device. 4A is output to the object data management device 2.

  In response to the search request 4A from the data search device 4, the object data management device 2 uses the data search unit 22 to index each object from the object data storage unit 23 in response to the search request 4A from the data search device 4. The desired object data is searched with reference to it. Then, the search result 4B including the obtained object data is returned to the data search device 4.

The search control unit 41 of the data search device 4 displays the search result received from the object data management device 2 on the screen, or outputs it to an external device or a recording medium as electronic data, or transmits it to a communication network.
Thus, desired object data is provided from the moving object DB system.

[Move prediction operation]
Next, with reference to FIG. 2, the movement prediction operation in the movement prediction apparatus according to the first embodiment of the present invention will be described. FIG. 2 is a flowchart showing a movement prediction process in the movement prediction apparatus according to the first embodiment of the present invention.

[Movement status]
First, a characteristic movement state that can be used for prediction of movement state movement will be described. The state referred to here is an event “when an object makes a certain movement, the same movement is performed thereafter”. If this state is used, the movement destination of the subsequent object can be predicted by analyzing the movement of the object and specifying the state.

There are two states: a state where the object hardly moves (stagnation) and a state where the object advances at a constant speed (straight forward). Hereinafter, these will be described.
Stagnation: A state of almost no movement. However, since there is an influence of position fluctuation described later, it is rare to stop completely.
Straight running: A state where the vehicle is moving in a constant direction at almost the same speed.

A state that does not correspond to any of these “stagnation” and “straight forward” states is referred to as a “random walk” state in the present invention.
Random walk: A so-called random walk in which both speed and direction of movement are indefinite. However, when observing actual data, there are many cases where only the direction of movement changes while the speed is within a certain range, not a complete random walk.

[Position fluctuation]
Next, position fluctuation will be described. In any measured trajectory data, there was a phenomenon that the position of the object slightly shifted in a direction unrelated to the moving direction while the object was stopped or moved. In the present invention, this is called “positional fluctuation”. This is considered to be caused by both the error due to the sensor to be measured and the fact that the object itself has a size (the center position measured by the sensor fluctuates). Which is the cause is almost impossible to distinguish on the data.

The fluctuation width of the data is substantially equal to a larger value among the measurement error of the sensor and the size (diameter) of the object. For example, if the sensor measurement error is 1ε and the object size is 2ε, the fluctuation width is 2ε. In predicting the position, it is necessary to consider the influence of this fluctuation. The fluctuation is represented by θ _p . This value is obtained by actual measurement.

[Formulation of state]
Next, state formulation will be described. First, the term “movement state” is formally defined. “The trajectory data λ _{t, −m} is in the movement state C” means that the trajectory data λ _{t, −m} from time t−m (where t and m are positive integers) to t satisfy the condition C. It means being in a state. That is, each state is specified by the condition C. The three movement states of “stagnation”, “straight forward”, and “random walk” described above are defined by corresponding conditions C _st , C _sw , and C _rw , respectively.

The stagnant state (st) condition C _st is defined as | p when the position vector at time i (where t−m ≦ i ≦ t: i is a positive integer) in the trajectory data λ _{t, −m} is p (i). (I) It can be defined as -p (t) | = 0. That is _, it is in a state of complete stop where the maximum speed of λ _{t, −m} is zero.
However, since the actual data has fluctuations as described above, it is rare that the speed is completely measured as zero. Therefore, in consideration of the influence of fluctuation theta _p, defining a stagnant state as follows.
C _st : | p (i) −p (t) | <θ _p
p (i) ελ _{t, −m}
That is, the object in the stagnant state, the condition that during the time t-m of t p (t) is not Na apart distance theta _p or more from. Thereby, it is possible to accurately determine that the moving object is in a stagnation state in consideration of the position fluctuation.

The straight-running state (sw) condition C _sw is the moving speed (vector value) of the trajectory data λ _{t, −m} at time i (0 ≦ i ≦ m) v (i) = p (i) −p (i− 1), it can be defined as v (i) -v (t). That is _, any movement vector of λ _{t, −m} is in a state where the difference from the velocity vector v (t) = p (t) −p (t−1) at time t is zero.
However, there is a fluctuation effect as in the stagnation state. For this reason, the straight traveling state including fluctuation is defined as follows.
C _sw : | v (i) −v (t) | <θ _p
v (i) = p (i) -p (i-1)
Thereby, it is possible to accurately determine that the moving object is in the straight traveling state in consideration of the position fluctuation.

  If the random walk state (rw) is not classified as either stagnant or straight, it can be determined as a random walk state. For this reason, it may be considered that the condition for specifying the random walking state is unnecessary. However, if the condition of the state is not set, the prediction function cannot be determined uniquely. Therefore, a condition is defined as a condition of a random walking state so that any moving object is satisfied.

Each point of the trajectory data λ _{t, -m} of the object whose end point is p (t) and the maximum speed is v _max is always included in a circle with a radius iv _max centered on p (t). Random walk condition C _rw is defined as C _rw : | v (i) | ≦ iv _max (t−m ≦ i ≦ t), where v _max is the maximum speed (scalar value) of λ _{t, −m.} A random walk state including fluctuation is defined as follows.
C _rw : | v (i) | ≦ v _max + θ _p
Thereby, it is possible to accurately determine that the moving object is in a random walking state in consideration of the position fluctuation.

[Prediction function]
Let R (j) be the expected movement range of the moving object indicated by the trajectory data λ _{t, + n} from time t to future time j = t + n (where t <j ≦ t + n: j, n is a positive integer). When λ _{t, −m} is in a certain state C, R (j) can be obtained from λ _{t, −m} using a conditional expression included in C. The range R (j) is a closed figure having an area, and is represented by, for example, a rectangle, a circle, or a set of these. R (j) can be expressed by a function having λ _{t, −m} and time j as arguments _, such as f (λ _{t, −m} , j) = R (j). This f is called a prediction function.

  In the conventional technique, f is defined as a function that returns point coordinates. However, it is not realistic to predict the position where a future object moves as position information with extremely high accuracy, that is, point coordinates. Therefore, in the present invention, f is defined as a function that returns a range.

In the stagnation state, since the condition C _st : | p (i) −p (t) | <θ _p , the range in which the object moves in the future R _st (j) = f _st (λ _{t, −m} , j ) Is also a circle with a radius θ _p centered on p (t). However, when creating an MBR (Minimum Bounding Rectangle), it is more convenient that the prediction range is a rectangle, so the minimum rectangle including this circle is actually set as the prediction range.
Thereby, it is possible to accurately calculate the movement prediction range when the moving object is in a stagnation state in consideration of the position fluctuation.

In the case of the straight traveling state, from the condition C _sw : | v (i) −v (t) | <θ _p , R _sw (j) = f _sw (λ _{t, −m} , j) is as shown in FIG. , and p (t) + (j- t) v (t) centered with radius theta _p circular area of the. FIG. 3 is an explanatory diagram showing a movement prediction range in a straight traveling state. The predicted area is circular, but in order to prepare an MBR as in the case of stagnation, the smallest rectangle including this circle is set as the actual prediction range.
Thereby, it is possible to accurately calculate the movement prediction range when the moving object is in the straight traveling state in consideration of the position fluctuation.

In the case of a random walking state, from the condition C _rw : | v (i) | ≦ v _max + θ _p , R _rw (j) = f _rw (t + n) is a radius centered on p (t) as shown in FIG. It becomes a circle of nv _max + θ _p . FIG. 4 is an explanatory diagram showing a movement prediction range in a random walk state. However, if this prediction function is used as it is, the prediction range obtained by the prediction function becomes very wide and the redundancy becomes high. Therefore, assuming that the moving object is a random walk, a method is introduced in which a region near the outer circumference of the circle, which is difficult for the moving object to reach, is removed from the predicted range.

In the case of a random walk, when an object moves jt times at a moving speed v _max or less, the probability ρ existing in a circle with a radius r (0 ≦ r ≦ nv _max ) from the starting point p (t) is It is given by the formula.
ρ = (r / (j−t) v _max ) ^(j−t)

That is, in order to predict the position of the moving object at time t + n within the probability ρ, the radius of the predicted circle may be θ _p + nv _max × n√p. For example, if ρ = 0.7 and π = 5, the radius of the circle is 0.93 × 5 v _max . However, in actual trajectory data, v _max tends to take an excessive value, and the prediction rate hardly decreases even if it is set to about ρ = 0.6 to 0.8. This ρ is preset to an appropriate value by the manager of the apparatus.
Accordingly, it is possible to accurately calculate the movement prediction range when the moving object is in a random walking state in consideration of the position fluctuation.

[Move prediction process]
Whether the movement prediction range indicating the movement destination at time t + n satisfies the conditions C _st , C _sw , and C _rw of the stagnation, straight travel, and random walk states described above with respect to the trajectory data λ _{t, −m} of the moving object. This is performed by applying a prediction function corresponding to a state satisfying the condition to the trajectory data. As described above, any moving object satisfies at least the condition C _rw . For this reason, the moving object is always specified as any state. If the object satisfies both the stagnation and straight travel conditions, it is predicted that the object is in a stagnation state. This is because R _st (j) has a smaller prediction range than R _sw (j).

A specific example of the movement prediction process will be described with reference to FIG. The movement prediction unit 31 of the movement prediction apparatus 3 executes the movement prediction process of FIG. 2 in response to the prediction request 2A from the object data management apparatus 2.
First, the movement prediction unit 31 acquires trajectory data λ _{t, −m} = {p (t−m),..., P (t)} from the prediction request 2A (step 100: data acquisition means), Each moving speed vector v (i) = p (i) −p (i−1) is calculated from the trajectory data (step 101: moving speed calculating means).

Next, stagnation trajectory data condition _{C st: | p (i)} -p (t) | Inspect satisfies <theta _p (step 102: the movement status determining means), is satisfied C _st (step 102: YES), the movement prediction range of the moving object at time j is the area within the circle of radius theta _p centered at p (t) (step 103: moving the expected range calculation means).

On the other hand, if C _st is not satisfied (step 102: NO), it is checked whether the trajectory data satisfies the straight traveling condition C _sw : | v (i) −v (t) | <θ _p (step 104: movement state specification) Means), if C _sw is satisfied (step 104: YES), the movement prediction range of the moving object at time j is a radius θ _p centered at p (t) + (j−t) v (t). The region is within a circle (step 105: movement prediction range calculation means).

If C _rw is not satisfied (step 104: NO), it is checked whether the trajectory data satisfies the random walking condition C _rw : | v (i) | ≦ v _max + θ _p (step 106: moving state specifying means) When C _rw is satisfied (step 106: YES), the movement prediction range of the moving object at time j is an area within a circle having a radius (j−t) v _max + θ _p centered at p (t). (Step 107: Movement prediction range calculation means).

  As described above, according to the present embodiment, the movement prediction device determines the moving state of the moving object by determining which of the conditions indicating the moving state of the moving object is stagnation, straight ahead, and random walking. In addition, since the movement prediction range of the moving object is calculated according to the movement state of the identified moving object, the movement prediction range is calculated considering the position fluctuation appropriately according to the movement state of the moving object. Even when the movement destination of an object is predicted using data including position fluctuation (measurement error), high prediction accuracy can be obtained with a relatively small amount of calculation.

[Second Embodiment]
Next, an index creation operation in the object data management apparatus 2 of the moving object DB system will be described.
As an index structure for each moving object, a tree structure based on R-Tree on the d + 1 dimension obtained by adding a time axis to a d-dimensional space is used, as in TPR-Tree and STP-Tree. In the moving object DB system, there are many forms in which the position data of moving objects tracked by sensors are collected all at once and added together to a database.

  For this reason, in this embodiment, unlike general R-Tree, only the position data of the current time is always added to the object data all at once. That is, the case where data of past time is added is not handled. Assuming this, since the MBR can be created after previously calculating the distance, speed difference, etc. between moving objects, an index more efficient than a general R-Tree can be created.

  A specific MBR (Minimum Bounding Rectangle) creation method will be described. First, the position vector p (t) of each moving object at time t at which the MBR is created is checked, and the distance between the moving objects is checked. Next, it is divided into a predetermined number of classes using a group average method which is one of clustering methods. By this clustering, moving objects that are relatively close to each other belong to the same class. Note that the number of moving objects belonging to one class is determined in advance. The position of each moving object included in the created class is checked, and this is set as the MBR of the leaf node. The MBR of the intermediate node is created after the MBR of the leaf node is created. Thereafter, the movement destination is predicted for each object, and a prediction MBR from time t + 1 to t + n is created.

  When the actual moving object deviates from the predicted MBR, the following is performed. If the deviation of the moving object occurs within a predetermined time interval u after the MBR is generated, it is dealt with only by changing the size of the corresponding MBR at the corresponding time to a size including the moving object. If this deviation occurs beyond the time interval u, the MBR is regenerated for all moving objects. u is a time interval determined in advance by the administrator of the apparatus.

  When an MBR is created in this way, an MBR with a much smaller area is created than when an R-Tree is created while adding moving objects one by one. When a moving object that did not exist at the time t is added thereafter, the addition is performed from the MBR on the root side to the leaf side in the same manner as in normal R-Tree or the like.

[Third Embodiment]
Next, a data search operation in the object data management apparatus 2 of the moving object DB system will be described with reference to FIGS. FIG. 5 is a top view showing a hierarchical structure of the index (MBR). FIG. 6 is a tree structure diagram showing a hierarchical structure of an index (MBR).
The search for object data is performed in the same manner as R-Tree (existing technology). As a specific method, the most basic method will be described.

  First, the created index (MBR) is hierarchized as shown in FIG. 6 for each time. As shown in FIG. 5, the upper layer MBR is created so as to completely include the lower layer MBR. Then, the position and size of the lower MBR group included in the MBR are recorded. This operation is repeated toward the upper layer. When the operation is performed in this manner, the MBR of the uppermost layer becomes a shape including the MBRs of all the lower layers.

  The data search operation is an operation for obtaining all objects in a specified range. First, it is determined whether or not the designated range is included in the MBR at the top layer of the created hierarchical structure. Here, when the designated range is not included in the shape of the MBR, it means that there is no moving object to be obtained. That is, the answer “no corresponding object” is returned.

  When the specified range is included in the MBR, it is checked whether there is an overlapping portion between the shape of the lower MBR included in the MBR and the specified range. If there is an overlapping range, the same range overlapping check is performed for the MBR having the overlapping portion and the lower layer MBR group included in the MBR. In this way, a list of MBRs having portions overlapping with the designated range is extracted from among the MBRs in the lowest layer. Finally, object data relating to objects included in the specified range among the objects included in the extracted MB: R is extracted from the data storage unit 23. This extracted object data is returned as a search result 4B.

It is a block diagram which shows the structure of the moving object DB system in which the movement prediction apparatus concerning the 1st Embodiment of this invention is used. It is a flowchart which shows the movement prediction process in the movement prediction apparatus concerning the 1st Embodiment of this invention. It is explanatory drawing which shows the movement prediction range in a straight drive state. It is explanatory drawing which shows the movement prediction range in a random walk state. It is a top view which shows the hierarchical structure of an index (MBR). It is a tree structure figure which shows the hierarchical structure of an index (MBR). It is explanatory drawing which shows the position prediction method by TRP-Tree.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Trajectory data input device, 11 ... Data acquisition part, 1A ... Trajectory data, 2 ... Object data management apparatus, 21 ... Data generation part, 22 ... Data search part, 23 ... Data storage part, 2A ... Prediction request, 2B ... Prediction result, 3 ... movement prediction device, 31 ... movement prediction unit, 4 ... data search device, 41 ... data search unit, 4A ... search request, 4B ... search result, 5 ... locus data, 6A ... search request, 6B ... search result.

Claims (4)

A trajectory that includes an arithmetic processing unit that performs arbitrary arithmetic processing on input data, and a storage unit that stores data used by the arithmetic processing unit, and includes position information of moving objects at each time by the arithmetic processing unit. A movement prediction device that calculates a movement prediction range of the moving object based on data,
In the arithmetic processing unit,
Data acquisition means for acquiring trajectory data of a moving object from the outside and storing it in the storage unit;
Each time i based on the position vector p (t−m) to p (t) of the moving object from time t−m (where t and m are positive integers) included in the trajectory data of the storage unit. A moving speed calculating means for calculating a moving speed vector v (i) = p (i) −p (i−1) in t−m ≦ i ≦ t (i is a positive integer) and storing it in the storage unit; ,
Stagnation of locus data before term memory unit said moving object, a moving condition specifying means for specifying a moving state of the moving object by determining whether meet one of the conditions shown straight, and each moving state of the random walk,
And a moving prediction range calculation means for calculating a movement prediction range of the moving object according to the moving state of the specific to said mobile object,
The movement state specifying means includes
Sensors that measure the trajectory data, including the trajectory data from the time t-m (where t and m are positive integers) to the position vector p (t-m) to p (t) of the moving object from the time t. Width of the position fluctuation θ _p of the moving object, which is the larger of the measurement error or the diameter of the moving object, is | p (i) −p (t) | <θ _p (where t−m ≦ i ≦ t: i is a positive integer), the moving object is determined to be in a stagnation state,
The moving speed vector v (t−m) to v (t) of the storage unit and the position fluctuation width θ _{p of the} moving object are | v (i) −v (t) | <θ _p (where t -M ≦ i ≦ t: i is a positive integer), the moving object is determined to be in a straight traveling state,
The moving speed vectors v (t−m) to v (t) of the storage unit, the maximum moving speed v _max of these moving speed vectors , and the position fluctuation width θ _{p of the} moving object are represented by | v (i) | If the condition of <v _max + θ _p (where t−m ≦ i ≦ t: i is a positive integer) is satisfied, the moving object is determined to be in a random walking state,
The movement prediction range calculation means includes
When the moving state specifying unit determines that the moving object is in the stagnation state, the movement prediction range of the moving object at time j (where t <j ≦ t + n: j, n is a positive integer) is at least p ( t) as a region in a circle with a radius θ _p centered at t) ,
When the moving state specifying unit determines that the moving object is in the straight traveling state, the movement prediction range of the moving object at time j (where t <j ≦ t + n: j, n is a positive integer) is at least p ( t) + (j−t) v (t) as a region within a circle with a radius θ _p
When the moving state specifying unit determines that the moving object is in the random walking state, the movement prediction range of the moving object at time j (where t <j ≦ t + n: j, n is a positive integer) is at least p ( A movement prediction apparatus, characterized in that a region within a circle having a radius (j−t) v _max + θ _p centered at t) is used. The movement prediction apparatus according to claim 1,
The movement prediction range calculation means sets the probability of being at a position of radius r from p (t) at time j to (r / (j−t) v _max ) (j−t) Prediction device. Based on trajectory data including position information of a moving object at each time by a movement prediction device having an arithmetic processing unit that performs arbitrary arithmetic processing on input data and a storage unit that stores data used in the arithmetic processing unit. A movement prediction method for calculating a movement prediction range of the moving object,
The arithmetic processing unit, a data acquisition step of storing into said storage unit to obtain the locus data of the moving object from the outside,
The arithmetic processing unit includes the moving object position vector p (tm) to p (t) from time t-m (where t and m are positive integers) included in the trajectory data of the storage unit. Based on the above, a moving speed vector v (i) = p (i) −p (i−1) at each time i (where t−m ≦ i ≦ t: i is a positive integer) is calculated and stored in the storage unit. A moving speed calculating step,
Moving state the processing unit, the trajectory data of the storage unit to identify the moving state of the moving object by determining whether meet one of the conditions shown stagnation, straight, and each moving state of the random walk of the moving object Specific steps,
The arithmetic processing unit, and a moving prediction range calculation step of calculating a movement prediction range of the moving object according to the moving state of the identified said moving object,
The moving state specifying step includes:
Sensors that measure the trajectory data, including the trajectory data from the time t-m (where t and m are positive integers) to the position vector p (t-m) to p (t) of the moving object from the time t. Width of the position fluctuation θ _p of the moving object, which is the larger of the measurement error or the diameter of the moving object, is | p (i) −p (t) | <θ _p (where t−m ≦ i ≦ t: i is a positive integer), the moving object is determined to be in a stagnation state,
The moving speed vector v (t−m) to v (t) of the storage unit and the position fluctuation width θ _{p of the} moving object are | v (i) −v (t) | <θ _p (where t -M ≦ i ≦ t: i is a positive integer), the moving object is determined to be in a straight traveling state,
The moving speed vectors v (t−m) to v (t) of the storage unit, the maximum moving speed v _max of these moving speed vectors , and the position fluctuation width θ _{p of the} moving object are represented by | v (i) | If the condition of <v _max + θ _p (where t−m ≦ i ≦ t: i is a positive integer) is satisfied, the moving object is determined to be in a random walking state,
The movement prediction range calculation step includes:
When the moving state specifying step determines that the moving object is in the stagnation state, the movement prediction range of the moving object at time j (where t <j ≦ t + n: j, n is a positive integer) is at least p ( t) as a region in a circle with a radius θ _p centered at t) ,
When the moving state specifying step determines that the moving object is in the straight traveling state, the movement prediction range of the moving object at time j (where t <j ≦ t + n: j, n is a positive integer) is at least p ( t) + (j−t) v (t) as a region within a circle with a radius θ _p
When the moving state specifying step determines that the moving object is in the random walking state, the movement prediction range of the moving object at time j (where t <j ≦ t + n: j, n is a positive integer) is at least p ( A movement prediction method, wherein a region within a circle having a radius (j−t) v _max + θ _p centered at t) is used. The program for functioning a computer as each part which comprises the movement prediction apparatus of Claim 1 or Claim 2.

Images