JPH10160831A - Multiple target tracking device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve a target movement element calculation accuracy without providing any special additional device by setting the region of a space determined by a prediction value according to (n) movement models to a correlation region and selecting detection data that cab be correlated to a tracking target and calculating the reliability. SOLUTION: A prediction equipment 54 calculates a prediction value of, for example, a target position and a speed according to (n) movement models of a plurality of targets. A correlator 56 sets the region of a space determined by an obtained prediction value to a correlation range and selects detection data that can be correlated to a track target, for example, from a target observation position and a target observation distance measurement change rate. A reliability calculator 57 calculates the reliability of a plurality of n movement models and selected detection data and calculates the estimation value by an acceleration vector estimator 59. Further, a target movement judging equipment 60 judges the movement state of a target from an obtained estimation value and obtains the smoothed value of, for example a target position and a speed by a smoothing equipment 62, and a smoothing error evaluation equipment 63 calculates the smoothing error covariance matrix.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention uses a target observation device typified by a radio sensor such as a radar or an optical sensor such as an infrared camera for a moving object such as an aircraft or a flying object. The present invention relates to a multi-target tracking device that tracks the movement of a plurality of targets by estimating the true value of the position and speed of the target based on signal detection results from a target and clutter other than the target.

[0002]

2. Description of the Related Art FIG. 13 shows, for example, IEEE TRANSA.
CTIONS ON AUTOMATIC CONTR
OL VOL. AC-23, AUGUST 1978,
P618-626 "Tracking Method
s in a Multitarget Enviro
"Probabilistic D
1 is a processing procedure showing a conventional multi-target tracking method indicated as “ata Association Filter”, FIG.
FIG. 4 is a configuration diagram of a conventional multi-target tracking apparatus corresponding to the conventional multi-target tracking method of FIG.

In FIG. 13, in the conventional multi-target tracking method, the initial value of the target position, the smoothed value of the velocity and the initial value of the smoothed error covariance matrix are calculated based on the result of the observation of the target position in step 1 based on the usual Kalman filter theory. Set, step 1
In step 5, for example, a target motion model is set by a constant velocity linear motion model. In step 16, the target position and velocity predicted values after one sampling from the current time are calculated by the constant velocity linear motion model. A prediction error covariance matrix is calculated by the constant velocity linear motion model in which the error of the predicted value by the rapid linear motion model is estimated, the target observation position is input as detection data as a signal detection result in step 6, and the tracking target is calculated in step 7. As detection data correlated with, select detection data existing in a certain area of the space determined using the prediction error covariance matrix by the constant velocity linear motion model around the target predicted position by the constant velocity linear motion model, In step 18, the reliability of whether or not the detection data selected in step 7 is detection data from the target to be tracked is converted into a constant velocity linear motion model. The gain is calculated using the predicted value and the prediction error covariance matrix based on the constant velocity linear motion model, the gain matrix used for smoothing the target motion parameters is calculated in step 9, and the target matrix using the constant velocity linear motion model is calculated in step 19. A smoothed value of the position and speed and a smoothed error covariance matrix based on the constant velocity linear motion model are calculated, and this series of steps is repeated until tracking is completed in step S14.

FIG. 14 is a block diagram of a conventional multi-target tracking apparatus. In FIG. 14, reference numeral 51 denotes a target which outputs a tracking target and a target observation position which is a signal detection result from a clutter other than the tracking target as detection data. The observation device 56 shares the prediction error calculated one sample before the current time by the constant velocity straight motion model with the target predicted position calculated by the constant velocity linear motion model one sample before the current time. A correlator for selecting detection data existing in a certain area of the space determined using a variance matrix, 58 a gain matrix calculator, 81 a predictor based on a constant velocity linear motion model,
82 uses a target predicted position calculated one sample before the current time by the constant velocity linear motion model and a prediction error covariance matrix calculated one sample before the current time by the constant velocity linear motion model; A detection data reliability calculator for calculating the reliability of whether or not the detection data selected by the correlator 56 is detection data from the target to be tracked, 83 is a smoother based on a constant velocity linear motion model, and 84 is a constant velocity Reference numeral 85 denotes a prediction error evaluator based on a constant speed linear motion model, and reference numerals 86 and 87 denote delay circuits.

The gain matrix calculator 58 calculates a gain matrix used for smoothing the target motion data from a prediction error covariance matrix calculated one sampling before the current time by a constant speed linear motion model. The predictor 81 based on the constant velocity linear motion model calculates the predicted value of the target position and the velocity after one sampling from the current time based on the smoothed value calculated by the smoother 83 based on the constant velocity linear motion model. I do. The smoother 83 based on the constant velocity linear motion model has a correlator 5
The detection data selected in step 6 and the reliability of the detection data calculated by the detection data reliability calculator 82 and the predicted value and gain matrix calculator calculated one sampling before the current time by the constant velocity linear motion model. The smoothed value of the target position and the speed is calculated using the gain matrix calculated in 58.

[0006] Next, a smoothing error evaluator 84 based on a constant velocity linear motion model is predicted by a predictor 81 based on a constant velocity linear motion model and is calculated by a gain matrix calculator 58 one sampling before the current time. Gain matrix, correlator 56
Detection data selected in the above, the reliability calculator 8 of the detection data
The evaluation value of the smoothing error of the target position and the speed is calculated using the reliability of the detection data calculated in step 2 and the prediction error covariance matrix calculated one sampling before the current time based on the constant velocity linear motion model. Further, the prediction error evaluator 85 based on the constant velocity linear motion model calculates an evaluation value of the target position and speed prediction error based on the smoothed error covariance matrix calculated by the smooth error evaluator 84 based on the constant velocity linear motion model. . The delay circuit 86 delays the predicted value calculated by the predictor 81 based on the constant speed linear motion model by one sampling. The replacement circuit 87 delays the prediction error covariance matrix calculated by the prediction error evaluator 85 based on the constant velocity linear motion model by one sampling delay.

[0007]

In the conventional multi-target tracking method and its apparatus as described above, the correlator 56 is used.
Is based on the predicted position of the tracking target calculated by the predictor 81 based on the constant velocity linear motion model assuming that the tracking target follows the constant velocity linear motion model, and the prediction error calculated by the prediction error evaluator 85 based on the constant velocity linear motion model. When the detection data input from the target observation device 51 exists in a certain area of the space obtained using the variance matrix, it is determined that the detection data is detection data that may have been detected from the tracking target. When the target makes a turning motion, the center point that determines the correlation range shifts greatly, or the size of the area of the space to be correlated is evaluated unduly small because the target turning situation is not considered. A problem occurred, and the tracking performance had to be degraded.

The present invention has been made in order to solve such a problem, and is intended for use in an environment in which a signal detection result such as a position and a distance change rate is obtained as detection data from a plurality of unnecessary signals such as a target and clutter. Another object of the present invention is to provide a multi-target tracking device that can accurately track a turning target.

[0009]

According to a first aspect of the present invention, a predicted value such as a target position and a speed is calculated by a target n motion models composed of a plurality of n constant vectors of the same dimension. A predictor based on n motion models and a region in the space defined by the predicted values based on the n motion models are used as a correlation range, and the possibility of correlation with a tracking target using a target observation position, a target observation distance change rate, and the like. A correlator for selecting detection data having a correlation; a plurality of n motion models; and a reliability calculator based on n motion models for calculating reliability of the detection data selected by the correlator; Constant vector estimator for calculating an estimated value of a constant vector constituting the motion model of the present invention, and constant vector estimation for determining a target motion state such as whether or not the target is turning based on the constant vector estimated value And a smoothed value such as a target position and a speed by using the n motion models while selectively switching the combination of the n motion models to be set based on the result of the judgment of the target motion state. And a smoothing error estimator based on n motion models for calculating a smoothing error covariance matrix estimating an error of the smoothed value.

According to a second aspect of the present invention, a predicted value such as a target position and a velocity is calculated by n target motion models composed of a plurality of n constant vectors of the same dimension.
A predictor based on the number of motion models and a space region determined by the predicted values based on the n number of motion models as a correlation range,
A correlator for selecting detection data that may be correlated with the tracking target using the target observation position, the target observation distance change rate, and the like,
A reliability calculator using the plurality of n motion models and the n motion models for calculating the reliability of the detection data selected by the correlator, and an estimated value of a constant vector forming the n motion models And a constant vector estimator that calculates the target distance change rate by using the n number of motion models, and then determines whether the target is turning based on the predicted value and the target distance change rate observation value. While selectively switching a combination of a target motion determiner based on a target distance change rate prediction value for determining a target motion state such as a target motion state and a set of n motion models based on a result of the determination of the target motion state, n Calculates smoothed values such as target position and velocity by using two motion models. Calculates a smoother using n motion models and a smoothed error covariance matrix that estimates an error of the smoothed values. It is provided with a smoothing error estimator according to number of motion models.

According to a third aspect of the present invention, a predicted value such as a target position and a speed is calculated by n target motion models composed of a plurality of n constant vectors of the same dimension.
A prediction error estimator based on n motion models, a prediction error evaluator based on n motion models for calculating a prediction error covariance matrix based on n motion models estimating an error of the predictor, and A correlator that selects a detection data that may be correlated with a tracking target using a target observation position, a target observation distance change rate, and the like, using a region of a space determined by the prediction error covariance matrix as a correlation range, and the plurality of n Number of motion models and a reliability calculator using n number of motion models for calculating the reliability of detection data selected by the correlator, and a constant for calculating an estimated value of a constant vector constituting the n number of motion models A vector estimator; a target motion determiner based on a constant vector estimate that determines a target motion state such as whether or not the target is turning based on the constant vector estimate; Based on the state of the determination result, sets n
While selectively switching combinations of individual motion models,
A smoother using n motion models for calculating smoothed values such as a target position and a velocity by using n motion models, and n motion models for calculating a smoothed error covariance matrix estimating an error of the smoothed values. And a smoothing error evaluator.

According to a fourth aspect of the present invention, a predicted value such as a target position, a velocity, etc., is calculated from n target motion models composed of a plurality of n constant vectors of the same dimension.
A prediction error estimator based on n motion models, a prediction error evaluator based on n motion models for calculating a prediction error covariance matrix based on n motion models estimating an error of the predicted value, and A correlator that selects a detection data that may be correlated with a tracking target using a target observation position, a target observation distance change rate, and the like, using a region of a space determined by the prediction error covariance matrix as a correlation range, and the plurality of n Number of motion models and a reliability calculator using n number of motion models for calculating the reliability of the detection data selected by the correlator, and a constant for calculating an estimated value of a constant vector constituting the n number of motion models After calculating a predicted value of the target distance change rate by using the vector estimator and the n number of motion models, the target is turning based on the predicted value and the target distance change rate observed value. A target motion determiner based on a target distance change rate predicted value that determines a target motion state such as whether or not, based on the determination result of the target motion state, while selectively switching a combination of n motion models to be set, A smoother based on n motion models for calculating smoothed values such as a target position and a velocity by using n motion models, and n motion models for calculating a smoothed error covariance matrix estimating an error of the smoothed values. And a smoothing error evaluator.

According to a fifth aspect of the present invention, a predicted value such as a target position and a speed is calculated by n target motion models composed of a plurality of n constant vectors of the same dimension.
A prediction error estimator based on n motion models, a prediction error evaluator based on n motion models for calculating a prediction error covariance matrix based on n motion models estimating an error of the predicted value, and The area of the space defined by the prediction value and the prediction error covariance matrix by the n motion models is used as the correlation range, and the detection data that may be correlated with the tracking target is calculated using the target observation position, target observation distance change rate, and the like. A correlator to be selected, and n for calculating the reliability of the plurality of n motion models and the detection data selected by the correlator
Number of motion models, a constant vector estimator for calculating the estimated value of the constant vector constituting the n number of motion models, and whether or not the target is turning based on the constant vector estimated value, etc. A target motion determiner based on a constant vector estimation value for determining a target motion state, and n number of motion models while selectively switching a combination of n motion models to be set based on the determination result of the target motion state. And a smoother estimator based on n motion models for calculating smoothed values such as a target position and a velocity by using a motion model and a smoothed error covariance matrix for estimating an error of the smoothed value. Are provided.

According to a sixth aspect of the present invention, there is provided a method for calculating a predicted value such as a target position, a speed, and the like based on n motion models of a target composed of a plurality of n constant vectors having the same dimension.
A prediction error estimator based on n motion models, a prediction error evaluator based on n motion models for calculating a prediction error covariance matrix based on n motion models estimating an error of the predicted value, and The area of the space defined by the prediction value and the prediction error covariance matrix by the n motion models is used as the correlation range, and the detection data that may be correlated with the tracking target is calculated using the target observation position, target observation distance change rate, and the like. A correlator to be selected, and n for calculating the reliability of the plurality of n motion models and the detection data selected by the correlator
Reliability calculator based on a plurality of motion models, a constant vector estimator for calculating an estimated value of a constant vector constituting the n motion models, and prediction of a target distance change rate by using the n motion models Calculating a target motion state such as whether or not the target is turning based on the predicted value and the target distance change rate observation value. Based on the determination result of the exercise state, a combination of the n exercise models to be set is selectively switched, and a smooth value such as a target position and a speed is calculated by using the n exercise models.
And a smoothing error estimator based on n motion models for calculating a smoothing error covariance matrix estimating an error of the smoothed value.

In the present invention, for example, a motion model obtained by adding n constant acceleration vectors including a zero vector to a constant velocity straight motion model as a plurality of n target motion models is used. The predicted value of the target position and the velocity is calculated using the reliability of each of the plurality of motion models and the constant acceleration vector constituting the plurality of motion models to calculate the influence term of the acceleration in the prediction. Calculated by adding the predicted value after one sampling calculated by the constant velocity straight-line motion prediction based on the smoothed value of Is used to calculate the parameters that increase the prediction error covariance matrix, which is the evaluation of the prediction error because the motion model is not uniquely determined. And calculates the prediction error covariance matrix by the n motion models by adding the prediction error covariance matrix for each number of motion models.

Further, in the present invention, when the target motion is stable and the motion model is uniquely determined as in the case of, for example, constant speed linear motion, the improper use of n motion models is In order to prevent poor tracking accuracy from affecting the rattling of the tracking, the motion state of the target is determined based on the estimated value of the calculated constant acceleration vector or the predicted value of the target distance change rate. If it is determined that the target is moving straight forward at a constant speed, the smoothed value and the smoothed error covariance matrix are calculated by using the n number of motion models. As a constant velocity straight-line model, all of the n constant vectors that are configured are replaced with zero vectors.
A smoothed value and a smoothed error covariance matrix are calculated.

According to the first and fourth aspects of the present invention, since a plurality of targets are tracked simultaneously and it is possible to cope with a case where detection data from clutter is large, an area to be correlated between the tracking targets overlaps. In order not to increase the detection data in the area to be correlated, and to increase the turning target coping ability while suppressing the increase in calculation load without expanding the area to be correlated, Use prediction values by n motion models to calculate points, use prediction error covariance matrices for each motion model instead of prediction error covariance matrices by n motion models to calculate the size of the area of space to be correlated Like that.

In the second and fifth aspects of the present invention, multi-target tracking by a target observation device having a long sampling interval and low observation accuracy in an environment where the number of tracking targets is small and the number of clutters is small is assumed, and target prediction is performed. When the acceleration term is added to the calculation of the value, in order to improve the turning target processing ability without causing the rattling of the tracking caused by the poor observation accuracy to surface, the calculation of the center point to be correlated takes n number of points. The size of the space area to be correlated using the predicted value of the constant velocity linear motion model when the constant acceleration vector, which is one of the predicted values for each motion model, is a zero vector, instead of the predicted value of the motion model A prediction error covariance matrix based on n motion models is used for the calculation.

In the third and sixth aspects of the present invention, extremely high-accuracy tracking is required, the accuracy of the target observation device is high, the sampling interval is short, and a multi-target system is used when a high-performance computer system is used. Assuming tracking, a prediction value based on n motion models is used to calculate a center point to be correlated, and a prediction error covariance matrix based on n motion models is used to calculate a size of a space region to be correlated. I have to.

[0020]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of a multi-target tracking apparatus according to the present invention will be described. FIGS. 1 and 2 show a processing procedure and a configuration, respectively, of a first embodiment of the multi-target tracking apparatus of the present invention. FIGS. 3 and 4 show a processing procedure and configuration of a second embodiment of the multi-target tracking apparatus of the present invention, respectively. FIG. FIG. 5 is a view showing a processing procedure for the third and fifth embodiments of the multi-target tracking apparatus of the present invention, FIG. 6 is a view showing a configuration of the third embodiment of the multi-target tracking apparatus of the present invention, and FIG. Fourth Embodiment of Target Tracking Apparatus 4, 6
FIG. 8 is a diagram showing a configuration of a multi-target tracking device according to a fourth embodiment of the present invention, FIG. 9 is a diagram showing a configuration of a multi-target tracking device according to a fifth embodiment of the present invention,
FIG. 10 is a diagram showing the configuration of a sixth embodiment of the multi-target tracking apparatus according to the present invention.

An embodiment of the present invention will now be described with reference to FIGS.
Prior to the explanation, the outline of the theory on which the present invention is based will be explained. In the following, symbols representing vectors are denoted by "vectors" in front of the symbols, such as vector X, in the code processing unit, and underlined in the image processing unit.

The target tracking processing start time by the multi-target tracking apparatus of the present invention is defined as t _1, and each sampling time at which the target tracking processing operation is performed in discrete time is represented by t _k (k = 1, 2, 2).
…). The sampling interval between the sampling time t _k−1 and the sampling time t _k is τ _k . That is, Equation (1) is used.

[0023]

(Equation 1)

A target motion model is defined as follows. That is, the target motion model at the sampling time t _k in the case where a plurality is n and equation (2). Hereinafter, symbols representing vectors are underlined. In Expression (2), a vector X _k is a state vector representing the true value of the target motion specification at the sampling time t _k , and as shown in Expression (3), represents a target position in rectangular coordinates and a target speed in rectangular coordinates. Let it be a vector as a component. Note that the symbol T represents transposition of a vector. Φ _k-1 is given by Expressions (4) and (5), assuming that the target performs a constant-velocity straight-line motion in a transition matrix of the state vector Χ _k from the sampling time t _k-1 to t _k .
The vector w _k is a driving noise vector at the sampling time t _k , and is a three-dimensional normal distribution white noise with an average vector 0, where E is a symbol representing the average, and is expressed by Equations (6) and (7).
Shall be satisfied. Here, Q _k is a driving noise covariance matrix at the sampling time t _k and is a value that does not depend on the motion model. Gamma _k is a transformation matrix driving noise vector at sampling time t _k, for example, the vector w _k Looking at the truncation error and the second term on the right side of equation (2) by the assumption that a constant speed rectilinear motion target motion model Is equivalent to an acceleration vector, and Γ _k is given by equation (8). Vector u _k
Is a constant acceleration vector forming n motion models at the sampling time t _k , which is represented by Expression (9), and Γ ′ _k
Is a transformation matrix of constant acceleration vector at sampling time t _k, the equation (10).

[0025]

(Equation 2)

FIG. 11 is a diagram for explaining the constant acceleration vector in a plane parallel to the plane. In the figure, O is the origin of coordinates O-xy with the target observation device as the origin, and X is x in the east direction.
The x-axis of the coordinate O-xy with the axis being positive, Y is the y-axis of the coordinate O-xy with the north being the positive y-axis, A1 is a constant acceleration vector in the positive direction of the y-axis, and A2 is the positive of the x-axis. A3 is a constant acceleration vector in the negative direction on the y-axis, and A3 is a constant acceleration vector in the negative direction on the y-axis.
4 is a constant acceleration vector in the negative direction of the x-axis. FIG.
Is 10 g (g is gravitational acceleration), and in the case of a motion model considering a constant acceleration vector of acceleration vector 0, n = 5, and equation (9) becomes As in 10), it can be written regardless of the sampling time.

[0027]

(Equation 3)

At sampling time t _k , equation (12)
Is written as Ψ _{k, L} (L = 1, 2,... N). Assuming Markov nature in the temporal transition of the motion model,
The motion model Ψ _{k, L} is determined by the motion model at the sampling time t _k−1 and does not depend on the motion model up to the sampling time t _k−2 . At this time, the transition probability P _LkLk-1 of the motion model is _represented by Expression (13).

[0029]

(Equation 4)

Next, an observation system model is defined as follows. Detection data from the tracking target goal and is at most one obtained at sampling time t _k, for the observation system model equation (14). Where the vector z _k is given by the equation (1)
5) to a target observation vector consists of the observed values of the target observation position and the target range rate by the orthogonal coordinates at sampling time t _k as shown, also F (vector x _k) is the formula (16), formula ( 17) is a non-linear function. The vector v _k is an observation noise vector corresponding to the target observation vector at the sampling time t _{k, and} is a four-dimensional normal white noise with an average vector 0, and satisfies the equations (18) and (19). Note that R _k is an observation noise covariance matrix at the sampling time t _k and is a value that does not depend on the motion model.

[0031]

(Equation 5)

Detection data from a target other than the target to be tracked,
In other words, it is assumed that the misdetection data is
Sampling time t_kFrequency per unit volume in
To λ ^k _FTAnd the area of space to be correlated with the tracking target
Volume to V_GkAnd the detection data from other than the tracking target
The total number of data in the area to be correlated is the average λ ^k
_FTV_GkAnd follow a Poisson distribution of Here, V_Gk
Is the volume of the four-dimensional space because the detection data is four-dimensional
It is.

The sampling time t the total number of detection data that has entered the area to take correlated with m _k pieces at _k, the vector z _{k, 1} in its entirety the detection data, vector z _{k, 2,} ..., vector z _{k , mk} and written as in equation (20). Further, information on the entire number of detected data from the sampling time t ₁ to t _k is written as M ^k, that is, equation (21)
The overall detection data from the sampling time t ₁ to t _k Z ^k That is, written as Equation (22).

[0034]

(Equation 6)

Next, some supplements are made to the target motion model. The conditional probability density function of the constant acceleration vector vector α _L (L = 1, 2,..., N) constituting the plurality of motion models is calculated according to the discrete value type probability theory according to equation (2).
3), Equation (24). Here, the delta function δ (vector x) has the property of Expression (25).

[0036]

(Equation 7)

From equations (24) and (25), equation (2)
6). That is, when Expression (27) is used, Expression (2)
8). Equation (29) is obtained from Equations (24) and (25).

[0038]

(Equation 8)

Next, the sampling time t_kDetect up to
At the time the data was obtained,
And sampling time t_kMultiple detection data obtained in
We define the reliability of the hypothesis. Vector z
_{k, i}(I = 1, 2,..., M_k) Is the search from the tracking target
The hypothesis that it is knowledge data_{k, i}(I = 1, 2,..., M
_k), And no detection data is obtained from the target to be tracked
The hypothesis χ_{k, o}Write Sampling time t_kFor up to
Conditional probability density based on detection data information
Expressions (30) to (33) are defined by the degree function.
You.

[0040]

(Equation 9)

Here, it is assumed that equations (34) and (35) are used. At this time, Expression (36) is obtained from Expression (13) and Expression (35). Equations (37) to (4) are based on probability theory.
5) is established.

[0042]

(Equation 10)

The following describes the target tracking processing procedure when the target motion model, observation system model, hypothesis and its reliability are defined as described above.

First, a method of smoothing and predicting under each hypothesis will be described. That is, based on the theory of the ordinary Kalman filter, when the extended Kalman filter is applied, the equation (4)
6) to Equation (55).

[0045]

[Equation 11]

Here, the vector x ＾^LK _k(-) Is a hypothesis.
_{k, lk}Sampling time t under _kVector for
x_kIf you write with the conditional mean vector
Equation (56) is based on the Kalman filter theory. Also,
Vector x ＾_k(-) Indicates the sampling time t_kAgainst
Vector x_kExpression (57) using the prediction vector of Be
Kuttle x ＾_k(+) Indicates sampling time t_kAgainst
Kutor x_kEquation (58) with the smoothed vector of Vect
X_{k, i, Lk}(+) Is a hypothesis._{k, i}And hypothesis Ψ _{k, lk}Nomo
And sampling time t_kVector x for_kFlat
This is expressed by equation (59) using a slip vector. Furthermore, P
_k(+) Indicates sampling time t_kError covariance at
Equation (60) is a scatter matrix. Also, P_k ^Lk(-) Is sun
Pulling time t _kHypothesis in Ψ_{k, lk}Mispredictions under
Expression (61) is a difference covariance matrix. P_{k, i, Lk}(+)
Sampling time t_kHypothesis in χ_{k, i}And Ψ_{k, lk}of
Equation (62) is the original smooth error covariance matrix.

[0047]

(Equation 12)

K _k is a gain matrix at the sampling time t _k . Further, the initial value vector x ＾ ₀ (+), P
₀ (+) is determined separately. Note that equation (47)
Since P ^Lk _k (−) is a value that does not depend on the hypothesis Ψ _{k, lk} , K _{k is obtained} from equation (48) and P _{k is obtained} from equation (52).
_{k, i, Lk} (+) (i = 1, 2,..., m _k ) are also values that do not depend on the hypothesis _{1 k, 1k} .

Next, a method of correlating the detection data with the tracking target will be described. When the detection data vector z satisfies Expression (63) using d as a parameter, it is determined that the detection data vector z has a correlation with the tracking target. Where the vector z
_k (−) is given by equation (64). Also, S _k is given by equation (6)
5). Here, P _k (−) is the prediction error covariance matrix at the sampling time t _k , and the equation (6)
6).

[0050]

(Equation 13)

FIG. 12 is a diagram for explaining the correlation between the detection data and the tracking target according to the equation (63) when the detection data has a two-dimensional dimension as a simple example. In the drawing, P indicates the observation from the tracking target. Vectors z _k (−) and Q in equation (64), which are predicted points, are calculated by linear algebra from parameter d and S _{k in} equation (65) at a boundary line that defines the inside and outside of the range to be correlated. D1 to D4 are detection data.

The target to be tracked is within the area to be correlated.
The probability of being P_GkWrite That is, Equation (67) is used.
Where region G_kIs represented by equation (68), and N (vector
Le z _k; A_k, A_k) Is the average vector a_k, Covariance matrix
A_kVector z of the four-dimensional normal distribution of_kProbability density at
Function. In addition, P_GkDepends on the value of d
It is decided intentionally.

[0053]

[Equation 14]

Next, a method of calculating the reliability of each hypothesis will be described.
Show. Hypothesis Ψ_{k, lk}Data vector z under the condition_{k, i}
Probability density function l_Lk(Vector Z_{k, i}) (L_k= 1
2, ... n; i = 1,2, ... m_k) Is close by 4D normal distribution
If it resembles, the equation (69) is based on the Kalman filter theory.
You. As can be seen from equation (69), the detection data vector
z _{k, i}Is the prediction vector F (vector x ＾^Lk _k
(-))_Lk(Vector z_{k, i}) Is large
And the vector z_{k, i}Is a hypothesis_{k, 1k}Luck under
Movement data and the detection data obtained from the tracking target
The reliability that there is is highly evaluated. Sampling time t
_kHypothesis at the time when detection information is not available
_{k-1, lk-1}And hypothesis Ψ_{k, 1k}Is true, the hypothesis Ψ
_{k-1, lk-1}Of equation (32), which is the reliability of^Lk-1 _k-1and
P in equation (13), which is the transition probability_LkLk-1From equation (70)
More. Also, the probability that the target is detected is P_DToss
If the target to be tracked exists within the area to be correlated,
The probability of being detected is P_DP_GkIt is. Therefore sample
Time t_kHypothesis at the time when the detection information was obtained.
_{k-1, lk-1}, Ψ _{k, lk}And hypothesis χ_{k, 0}Confidence that is correct
Is the sampling time t_kInformation is not available
Expression (70), which is the reliability at the time of
Probability that detection data will not be obtained from within the area to be correlated
lp_DP_GkAnd m_kAll of the detection data are tracking pairs
Reliability of detection data from objects other than the target
(Λ^k _PT)^mk, Ie, proportional to equation (71)
You may think that.

The hypothesis _{{k-1, lk-1 , Ψk, lk} and hypothesis _} at the time when the detection information at the sampling time t _k is obtained.
The reliability that _{k, i} (i = 1, 2,..., m _k ) is correct is
In equation (70), which is the reliability at the time when the detection information is not obtained at the sampling time t _k , the probability P _D that the tracking target is detected, the detection data vector z _k under the hypothesis Ψ _{k, lk ,} evaluation value l _Lk (vector z _{k, i)} of _i and m _k -
It may be considered that it is proportional to the value obtained by multiplying the reliability (λ ^k _PT ) ^mk−1 that one piece of detection data is obtained from a target other than the tracking target, that is, equation (72). Therefore, equation (3)
Β _{k, i, Lk, L k−1} of (0) is _obtained by using equation (45) and equation (7).
1) and Expression (72) are normalized to obtain Expressions (73) to (7).
It is calculated as 5). Note that β _{k, i} in Expression (31), β ^Lk _{k in} Expression (32), and β _{k, i, Lk in} Expression (33) are _{obtained by} Expression (73).
To ( _{k), i, Lk , and Lk-1} calculated in Equations (75) to (37), (39), and (41), respectively.

[0056]

(Equation 15)

Next, a method of calculating the estimated value of the acceleration vector will be described. The estimated value of the acceleration vector at the time of detection data to the sampling time t _k is obtained the formula (7
Defined in 6). Expression (77) is obtained from Expression (76) in the same manner as Expressions (27) and (28). Therefore, Expression (78) is obtained by applying Expression (32) to Expression (77).

[0058]

(Equation 16)

The smoothed vector of the target position and the velocity under all the hypotheses, that is, the vector x ＾ of the equation (58)
_k (+) is calculated by integrating the smoothed vectors found under each hypothesis using the reliability of each hypothesis. First, from equation (33) and Bayes' theorem, equation (7)
9) is obtained. Equation (58), Equation (59), Equation (79)
Equation (80) follows. Equation (80) and Equation (49),
From Expression (51), Expression (46), Expression (78), Expression (42), and Expression (44), Expressions (81) to (83) are obtained.

[0060]

[Equation 17]

The target position under all the above hypotheses,
The smoothing error covariance matrix of the velocity smooth vector, that is, P _k (+) in equation (60) is calculated as follows. That is, first, Expression (84) is obtained from Expression (60), Expression (58), Expression (59), Expression (62), and Expression (79). From equation (42) and equation (80), equation (85) gives
From Expressions (84) and (85), Expression (86) is obtained. In Expression (86), Expression (50), Expression (52), Expression (38), Expression (43), Expression (49), Expression (51), Expression (46), Expression (78), Expression (81)- Equation (83) and Equation (42),
Substituting equation (44), P ^Lk _k (-) and P
Note that _{k, i, Lk} (+) is independent of _{Ψk, lk, yielding} equation (87).

[0062]

(Equation 18)

The predicted vector of the target position and the velocity based on the smoothed vector under all the above-mentioned hypotheses, that is, the vector x ＾ _k (−) and the vector x ＾ in the equation (57)
The prediction error covariance matrix of _k (−), that is, P _k (−) in equation (66) is calculated as follows. First, with respect to the predicted vector x ＾ _k (−), the equation (2) is substituted into the equation (57), and the equation (88) is obtained by applying the equations (58), (6) and (27). Here, since equation (89) is obtained according to probability theory from equation (70), equations (28) and (8)
Equation (90) is obtained from 9). Next, the prediction error covariance matrix P
_{Since k} (−) is equation (91) from equations (2) and (88), equation (91) is substituted into equation (66), and equations (60), (7), and (29) ) Is applied, and if the vector x _k−1 , the vector w _k−1 , and the vector u _k−1 are not correlated with each other, the equation (92) is obtained. Therefore, from Expression (92), Expression (47) and Expression (89), Expression (93) is obtained.
It is.

[0064]

[Equation 19]

In the above-described multi-target tracking method, as a special case, all the n constant acceleration vectors in the equation (12) are set to zero vectors, that is, the equation (94)
The method includes a multi-target tracking method using a constant velocity linear motion model. When Equation (94) is used, Equation (94) is applied to Equations (46) and (78), and Equation (4) is applied to the equations for calculating the smooth vector and the smooth error covariance matrix.
By using 3), equations (81) to (83) become equations (95) to (97), and equation (87) becomes equation (9).
8).

[0066]

(Equation 20)

Next, a description will be given of a method of judging whether the target motion is going straight or turning on the basis of various amounts calculated in the process of the multi-target tracking method. Sampling time t
_The result of the determination of whether the target movement is straight or curved at _k is determined by the flag F
Expressed in MD _k, the determination result is expression if the straight (98), the determination result is defined as the case of curvilinear progression formula (99). The initial value of the determination result flag at the tracking start time is set to 0 (straight).

[0068]

(Equation 21)

First, a method of judging whether the target movement is straight or curved based on the estimated acceleration vector will be described. The x-, y-, and z-components of the estimated acceleration vector u _k-1 in equation (78) are respectively estimated as u _k-1 (X), u
Write as _k-1 (y) and u _k-1 (z) to obtain the equation (100). A vector estimated value _obtained by smoothing the acceleration vector estimated value u _k-1 by a primary filter for each component is defined as u ^s _k-1, and its x, y, and z components are estimated values u ^s respectively.
_{k-1 (x), u} s k-1 (y), write u ^s _k-1 and (z).
That is, equations (101) to (104) are used. here,
μ is the gain of the primary filter, and has the same value for each of the x, y, and z components. Moreover, estimates in the tracking start time _{^{u s k-1 (x)}} , is set to _{^{u s k-1 (y)}} , u s k-1 all initial value 0 (z). Euclidean norm estimate of the vector estimate ^{_{u s k-1 ‖u s k}} -1 ‖ formula (105)
Becomes

[0070] straight / curvilinear progression of the desired motion at the sampling time t _k is determined in the following manner. That is, equation (10)
If 6) is satisfied, the target exercise is determined to be straight ahead, and the equation (9) is used.
8). When Expression (107) is satisfied, the target motion is determined to be curved, and Expression (99) is obtained. When Expression (108) is satisfied, the determination result at the sampling time t _k−1 is held, and is set as Expression (109). Here, ζ ₁ and ζ ₂ are constants that do not depend on the sampling time and satisfy Expression (110).

[0071]

(Equation 22)

Second, based on the predicted value of the target distance change rate,
A method of determining whether the target exercise is going straight or turning will be described. R. ^P _k
If defined by a formula to the predicted value of the target range rate for the sampling time t _k (112), an expression (68) from the equation using the equation (17) (113). Equation (65)
The fourth row fourth column element of the S _k and S _k (4, 4). That is, Expression (114) is used. The detection data vector z _{k, i} (i = 1, 2,..., M) at the sampling time t _k
Each target range rate component of _{^{k) R. o k, i}} (i =
1,2, ..., a m _k), if the approximation ^{R. O} _k at the time the detection data to the sampling time t _k is _obtained, the probability density function of _i in one-dimensional normal distribution of formula (115) , the target range rate is defined by the formula _{^{(115) R. o k,}} i (i
= 1,2, ..., weighted average probability density function of the estimated value ^{R. O} _k of m _k) is using equation (38), also _{^{R. O k, i (i}} =
1, 2,..., M _k ) are independent of each other,
17) becomes a one-dimensional normal distribution. Therefore, if the scalar quantity ε _k is defined by equation (118), ε _k follows a chi-square distribution with one degree of freedom.

The straight / turning of the target motion at the sampling time is determined as follows. That is, equation (119)
Is satisfied, the target exercise is determined to be straight ahead, and equation (99) is set. When Expression (120) is satisfied, the target motion is determined to be a curve, and Expression (100) is obtained. When Expression (121) is satisfied, the determination result at the sampling time t _k-1 is held.
That is, Expression (109) is used. Here, η ₁ and η ₂ are constants that do not depend on the sampling time and satisfy Expression (121).

[0074]

(Equation 23)

Next, Embodiment 1 of the present invention will be described with reference to FIGS.
It is explained according to. Based on the information of the target position and the target distance change rate obtained from the target observation device 51, based on the theory of the ordinary Kalman filter, the initial values of the target position, the smoothed value of the velocity, and the smoothed error covariance matrix are set (step 1). ), N target motion models composed of n constant acceleration vectors of the equation (12) including the zero acceleration vector as in the example of the equation (11) are set (step 2), and tracking is continued. Into a processing loop to perform

Hereinafter, the flow of processing in the loop will be described. Explanation
For convenience, the sampling time at the beginning of the loop is t
_k-1It is assumed that Predictor 52 for each motion model
At the input from the smoother 62 with n motion models
Value vector by n motion models of the equation (58)
Torr x ＾_k-1Construct (+) and n motion models
The constant acceleration vector α in equation (12)_LkEquation (4)
6) According to the motion model prediction value vector x ベ ク ト ル^Lk
_kCalculate (-) and evaluate the prediction error for each motion model
Error evaluation using the n motion models
Error by the n motion models input from the unit 63
Covariance matrix P_k-1(+) And a preset expression
(7) driving noise covariance matrix Q_k-1Equation (47)
Prediction error covariance matrix P for each motion model
^Lk _k(−) Is calculated (step 3). Then n exercises
In the model predictor 54, n motion models
(58) from the smoother 62
Smoothed value vector x ＾_K-1Enter (+) and go straight at constant speed
Φ_k-1Vector x ＾_K-1(+)
Obtained from the reliability calculator 57 based on n motion models
The reliability β of each motion model in equation (32)^Lk-1 _k-1,Araka
Transition probability P of equation (13) set in advance_LkLk-1And expressions
(12) Constant acceleration vector α_LkEquation (90)
The sampling time t _kInformation at
Estimating the influence term of the acceleration at the time when there is no
Accordingly, a predicted value vector x ＾ by n motion models_k
(−) Is calculated (step 4).

Next, after the time has elapsed from the sampling time t _{k −1} to the sampling time t _k , the target observation device 51
To enter more positions and range rate from the target and clutter like detection result as detection data sampling time t _k (Step 6). Next, in the correlator 56, the predicted value vector x ＾ _k (−) based on the n motion models of Expression (57) calculated at the sampling time t _k−1 is passed through the first delay circuit 64 to the n motion models. (64), which is a central position to be correlated by input from the predictor 54
Vector z _k (-) was calculated by the equation (16), the prediction error covariance matrix P ^Lk _k for each motion model of equation (61) calculated in sampling time (-) the second delay circuit 65
Through the prediction error evaluator 53 for each motion model, and through the first delay circuit 64 and the observation noise covariance matrix R _{k of the} preset equation (19), the prediction by the n motion models is performed. From the vector x _k (−) input from the unit 54, S _k of the equation (65) used for calculating the size of the area to be correlated is replaced by P ^Lk instead of P _k (−).
_k (-) determined by using, select the m _k pieces of detection data of possible correlation with the tracking target goal by equation (63), which is referred to as Z _k of the formula (20) (Step 7 ).

Next, a reliability calculator based on n motion models
At 57, the sampling time t _k-1To equation (39)
Reliability β of each motion model calculated from^Lk-1 _k-1The third
Input through the delay circuit 66, and the sampling time t_k-1
Prediction vector for each motion model of equation (56)
X^Lk _k(-) Is passed through the fourth delay circuit 67 and the motion model
Input from the predictor 52 for each
_k-1Error calculated by equation (61) for each motion model
Covariance matrix P^Lk _k(−) Through the second delay circuit 65
Input from the prediction error evaluator 53 for each motion model
The observation noise covariance matrix R of equation (19) set in advance
_kFrom Expression (69), Expression (16), Expression (17) and Expression
(53), hypothesis に よ り_{k, lk}Correlation under
Data vector z from the detector 56_{k, i}L
_Lk(Vector z_{k, i}), And a preset search
Knowledge probability P_DThe target to be tracked in equation (67) should be correlated.
Probability P in the region_GkAnd the motion model of equation (13)
Dell transition probability P_LkLK-1Equations (73) to (75)
And according to equations (39), (41) and (43)
Motion model and detection data reliability β_{k, i, Lk, Lk-1}β
^Lk _k, Β_{k, i, Lk}, Β_{k, i}Is calculated (step 8).

Next, the gain matrix calculator 58
Pulling time t_k-1(61) motion model calculated
Prediction error covariance matrix P^Lk _k(-) Is the second delay time
From the prediction error evaluator 53 for each motion model through the road 65
Input and sampling time t _k-1Equation (57) calculated to
X ＾ taken to be the predicted value by the n motion models of_k(-)
Is passed through the first delay circuit 64 and the prediction by n motion models is performed.
Equation (19) input from the measuring instrument 54 and set in advance
The observation noise covariance matrix R_kFrom equation (48) and equation (5)
3) calculating a gain matrix according to equation (54)
Step 9). Next, in the acceleration vector estimator 59,
Equation (3) is obtained from the reliability calculator 57 using n motion models.
2) β^Lk _kIs input and the preset expression (1
2) Constant acceleration vector vector α_LkAnd the equation (7)
8) Calculate the acceleration vector estimated value of equation (76) according to
Issue (Step 10).

Next, the target motion based on the estimated value of the acceleration vector
In the decision unit 60, the expression
(76) acceleration vector estimated value u_k-1Enter the goal
Sampling time t obtained by motion determiner 60_k-1In
Motion judgment result FMD _k-1To the fifth delay circuit 68
And input according to equations (103) to (105).
Vector estimate u of the acceleration vector estimate^S _k-1
Is calculated, and the acceleration vector is calculated according to the equation (106).
Vector estimate u of the estimated value^S _k-1Euclid's
Norm estimate ‖u^s _k-1After calculating ‖, determine the target movement
Result FMD_kIf the expression (107) holds, the expression
(99), and equation (108) holds.
In such a case, it is set as in equation (100), and equation (10)
If 9) is established, set as in equation (110)
(Step 11).

Next, in the smoother 62 based on the n motion models, the smoothed value vector x ＾ based on the n motion models of the equation (58) calculated at the sampling time t _k−1 is obtained.
_k−1 (+) is input through the sixth delay circuit 69, and the reliability β ^Lk _k of each motion model of Expression (32) and Expression (3)
The detection data vector z _{k, i} under the hypothesis Ψ _{k, lk} of 3)
Confidence beta _{k, i,} and input from the reliability calculator 57 _Lk by n number of motion models, the correlator 56 the detection data vector z _{k, 1} from the vector z _{k, 2,} ..., vector z _{k, mk} Is input and the equation (5) calculated at the sampling time t _k−1 is input.
6) The predicted value x ^1k _k (−) for each motion model is input from the predictor 52 for each motion model through the fourth delay circuit 67, and the gain matrix K _k of equation (48) is calculated by the gain matrix calculator 5
8, the acceleration vector estimated value u of equation (76)
_k-1 is input from the acceleration vector estimator 59, the target motion determination result FMD _k is input from the target motion determiner 60 based on the estimated acceleration vector, and the target motion determination result FMD _k is 1 (curvature). Is a smoothed vector x ＾ _k based on n motion models according to equations (81) to (83).
(+) Is calculated, and if the target motion determination result FMD _k is 0 (straight ahead), a smoothed value vector x based on n motion models is obtained according to Expressions (95), (96), and (97). ＾ Calculate _k (+).

Smooth error estimator 6 based on n motion models
3, a reliability calculator 57 based on n motion models
More type beta _{k, i} of formula (31), beta _{k, i} of beta ^Lk _k of the formula (32) _(33), the _Lk, sampling time t _k-1
The prediction error covariance matrix P ^Lk _k (−) for each motion model calculated in equation (61) is input from the prediction error evaluator 53 for each motion model through the second delay circuit 65, and the gain matrix calculator 58 From Equation (48), a gain matrix K _k is input, and
The predicted value vector x ^Lk _k (−) for each motion model calculated in equation (56) calculated at the sampling time t _k−1 is input from the predictor 52 for each motion model through the fourth delay circuit 67, The predicted value vector x ＾ _k (−) based on the n motion models calculated at the sampling time t _k−1 is input from the predictor 54 based on the n motion models through the first delay circuit 64, and is input from the correlator 56. The detection data vector z _{k, 1} , vector z _{k, 2} ,..., Vector z _{k, mk} are input, and equation (76)
Acceleration vector estimate u _k-1 inputted from the acceleration vector estimator 59, the desired motion determination result FMD _k inputted from the desired motion determiner 60 based on the acceleration vector estimate, the desired motion determination result FMD _k is 1 (songs Hex)
A smooth error covariance matrix P _k (+) based on the n motion models of equation (60) is calculated according to equation (87) and equations (82) and (83), and the target motion determination result FMD _k is 0.
(Straight ahead), Equation (98) and Equation (9)
6) According to equation (97), a smooth error covariance matrix P _k (+) based on n motion models of equation (60) is calculated (step 13). This series of flows in the loop is repeated until tracking ends (step 14).

Next, a second embodiment of the present invention will be described with reference to FIGS.
It is explained according to. Based on the information of the target position and the target distance change rate obtained from the target observation device 51, based on the theory of the ordinary Kalman filter, the initial values of the target position, the smoothed value of the velocity, and the smoothed error covariance matrix are set (step 1). ), N target motion models composed of n constant acceleration vectors of the equation (12) including the zero acceleration vector as in the example of the equation (11) are set (step 2), and tracking is continued. Into a processing loop to perform

Hereinafter, the flow of processing in the loop will be described. For convenience of explanation, the sampling time at the beginning of the loop is t
_Assume that it was _k-1 . Predictor 52 for each motion model
, The smoothed value vector x ＾ _k-1 (+) based on the n motion models of the equation (58) input from the smoother 62 based on the n motion models and the equation (n) constituting the n motion models From the constant acceleration vector α _{Lk of} (12), the equation (4)
6) According to the motion model, the predicted value vector x ＾ ^Lk
_k (−) is calculated, and the prediction error evaluator 53 for each motion model outputs a smooth error covariance matrix P _k−1 based on n motion models input from the smooth error evaluator 63 based on n motion models. (+) And the driving noise covariance matrix Q _{k-1 of the} previously set equation (7), the prediction error covariance matrix P for each motion model according to the equation (47).
^Lk _k (−) is calculated (step 3).

Next, in a predictor 54 based on n motion models, a smoother 62 based on n motion models obtains the formula (5)
8) Smooth value vector x by n motion models
_k−1 (+) is input, a Φ _k−1 vector x ＾ _K−1 (+) is calculated by constant-velocity straight-line motion prediction, and the equation (32) obtained from the reliability calculator 57 using n motion models is obtained. ), The reliability β ^Lk-1 _k-1 of each motion model, and a preset equation (1
According to the transition probability P _{LkLk-1 of} 3) and the constant acceleration vector α _Lk of Expression (12), the influence term of the acceleration at the time when the detection information is not obtained at the sampling time t _k is estimated according to Expression (90), According to equation (88), a predicted value vector x ＾ _k (−) based on n motion models is calculated (step 4).

Next, after the time has elapsed from the sampling time t _{k -1} to the sampling time t _k , the target observation device 51
To enter more positions and range rate from the target and clutter like detection result as detection data sampling time t _k (Step 6).

Next, in the correlator 56, the prediction value vector x ＾ _K (−) based on the n motion models of the equation (57) calculated at the sampling time t _k−1 is passed through the first delay circuit 64 to the n Input from the predictor 54 based on the motion model of
The vector z in equation (64) that is the center position to be correlated
_k (−) is calculated by equation (16), and the prediction error covariance matrix P ^Lk _k (−) for each motion model of equation (61) calculated at the sampling time is passed through the second delay circuit 65 for each motion model. From the prediction error evaluator 53, and the observation noise covariance matrix R of the equation (19) set in advance.
_k and a vector x ＾ input from a predictor 54 based on n motion models through the first delay circuit 64
_{According to k} (−), S in equation (65) used for calculating the size of the area to be correlated is replaced by P _k (−) instead of P _k (−).
^Lk _k (-) determined by using, select the m _k pieces of detection data of possible correlation with the tracking target goal by equation (63), which is referred to as Z _k of the formula (20) (step 7).

Next, a reliability calculator based on n motion models
At 57, the sampling time t _k-1To equation (39)
Reliability β of each motion model calculated from^Lk-1 _k-1The third
Input through the delay circuit 66, and the sampling time t_k-1
Prediction vector for each motion model of equation (56)
X^Lk _k(-) Is passed through the fourth delay circuit 67 and the motion model
Input from the predictor 52 for each
_k-1Error calculated by equation (61) for each motion model
Covariance matrix P^Lk _k(−) Through the second delay circuit 65
Input from the prediction error evaluator 53 for each motion model
The observation noise covariance matrix R of equation (19) set in advance
_kFrom Expression (69), Expression (16), Expression (17) and Expression
(53), hypothesis に よ り_{k, lk}Correlation under
Data vector z from the detector 56_{k, i}L
_Lk(Vector z_{k, i}), And a preset search
Knowledge probability P_DThe target to be tracked in equation (67) should be correlated.
Probability P in the region_GkAnd the motion model of equation (13)
Dell transition probability P_LkLk-1Equations (73) to (75)
And according to equations (39), (41) and (43)
Motion model and detection data reliability β_{k, i, Lk, Lk-1},
β^Lk _k, Β_{k, i, Lk}, Β_{k, i}Is calculated (step 8).

Next, the gain matrix calculator 58
Pulling time t_k-1(61) motion model calculated
Prediction error covariance matrix P^Lk _k(-) Is the second delay time
From the prediction error evaluator 53 for each motion model through the road 65
Input and sampling time t _k-1Equation (57) calculated to
Predicted value vector x ＾ by n motion models_k(-)
Is passed through the first delay circuit 64 and the prediction by n motion models is performed.
Equation (19) input from the measuring instrument 54 and set in advance
The observation noise covariance matrix R_kFrom equation (48) and equation (5)
3) calculating a gain matrix according to equation (54)
Step 9).

Next, in the acceleration vector estimator 59,
Equation (3) is obtained from the reliability calculator 57 using n motion models.
2) β ^Lk _k is input, and the preset equation (1)
Using the constant acceleration vector α _Lk of 2), an acceleration vector estimated value of Expression (76) is calculated according to Expression (78) (Step 10).

Next, in the target motion determining unit 61 based on the predicted value of the target distance change rate, the predicted value vector x ＾ _k (−) based on the n motion models of the equation (57) calculated at the sampling time t _k−1 is calculated. Through a first delay circuit 64, the motion data is input from a predictor 54 based on n motion models, and the reliability β _{k, i} of the detected data is input from a reliability calculator 57 based on n motion models. The target motion determination result FMD _k-1 at the sampling time t _k-1 obtained in step 61 is the fifth
Enter through the delay circuit 68, S _k enter a further correlator 56 from the detection data vector z _k of the formula (65) used for the size calculation of the area to correlate the detection data from the correlator 56 _{, 1,} vector z _{k, 2,} ..., vector z _k, enter the _mk the range rate component ^{R. O} _{k, 1,
^{R. O k, 2, ...}} , R. O k, and _mk, equation (113), the predicted value of the target range rate by the formula (17) ^{R. P} _k, the fourth row of S _k by equation (114) The fourth column element S _k (4,
4) After the weighted average estimate of the target range rate ^{R. O} _k a respectively determined, further calculates a scalar quantity epsilon _k according to equation (118) by the equation (116), the desired motion judgment result F
When MD (119) is satisfied, MD _k is calculated by Expression (99).
And if equation (120) holds, set as equation (100), and if equation (121) holds, set as equation (110) (step 12) ).

Next, in the smoother 62 based on the n motion models, the smoothed value vector x ＾ based on the n motion models of the equation (58) calculated at the sampling time t _k−1 is obtained.
_k−1 (+) is input through the sixth delay circuit 69, and the reliability β ^Lk _k of each motion model of Expression (32) and Expression (3)
The detection data vector z _{k, i} under the hypothesis Ψ _{k, lk} of 3)
_Are input from the reliability calculator 57 based on n motion models, and the correlator 56 detects the detection data vector z _{k, 1} , vector z _{k, 2} ,..., Vector z _{k, mk} Is input and the equation (5) calculated at the sampling time t _k−1 is input.
6) Prediction vector x ＾ ^Lk _k (−) for each motion model
Is passed through a fourth delay circuit 67 and the predictor 5 for each motion model
2, the gain matrix K _k in equation (48) is input from the gain matrix calculator 58, and the estimated acceleration vector u _k-1 in equation (76) is input from the acceleration vector estimator 59,
The desired motion determination result FMD _k inputted from the desired motion determiner 61 according to the predicted value of the target range rate, if desired motion determination result FMD _k is 1 (curvilinear progression) have the formula (81) to (83 ), A smoothed value vector x ＾ _k (+) based on n motion models is calculated, and the target motion determination result FMD _k
Is 0 (straight ahead), Equation (95) and Equation (9)
6) According to equation (97), a smoothed value vector x） _k (+) based on n motion models is calculated.

[0093] In the smoothing error estimator 63 according to the n motion models, beta _{k, i} of formula (31) from the reliability calculator 57 by the n motion models, beta ^Lk _k of the formula (32) ( 33) Input β _{k, i, Lk} of sampling time t
The prediction error covariance matrix P ^Lk _k (−) for each motion model of Equation (61) calculated as _k−1 is input from the prediction error evaluator 53 for each motion model through the second delay circuit 65, and the gain matrix The gain matrix K _k of equation (48) is input from the calculator 58, and the equation (5) calculated at the sampling time t _k−1 is input.
6) Prediction vector x ＾ ^Lk _k (−) for each motion model
Is passed through a fourth delay circuit 67 and the predictor 5 for each motion model
2 and the predicted value vector x ＾ _k (−) based on the n motion models calculated at the sampling time t _k−1 is the first
Predictor 5 based on n motion models through delay circuit 64
4 and the detection data vector z from the correlator 56.
_{k, 1} , vector z _{k, 2} , ..., vector z _{k, mk} ,
The acceleration vector estimate u _k-1 of the formula (76) is input from the acceleration vector estimator 59, the desired motion determination result FMD _k
Was input from the desired motion determiner 61 according to the predicted value of the target range rate, if desired motion determination result FMD _k is 1 (curvilinear progression) have the formula (87) and (82), formula (83)
Calculates a smooth error covariance matrix P _k (+) based on the n motion models of the equation (60) according to the equation (60), and calculates the target motion determination result F
If MD _k is 0 (straight), the smooth error covariance matrix P _k (+) based on the n motion models of equation (60) according to equations (98), (96), and (97) Is calculated (step 13). This series of flows in the loop is repeated until tracking ends (step 14).

Next, a third embodiment of the present invention will be described with reference to FIGS.
It is explained according to. Based on the information of the target position and the target distance change rate obtained from the target observation device 51, based on the theory of the ordinary Kalman filter, the initial values of the target position, the smoothed value of the velocity, and the smoothed error covariance matrix are set (step 1). ), N target motion models composed of n constant acceleration vectors of the equation (12) including the zero acceleration vector as in the example of the equation (11) are set (step 2), and tracking is continued. Into a processing loop to perform

Hereinafter, the flow of processing in the loop will be described. For convenience of explanation, the sampling time at the beginning of the loop is t
_Assume that it was _k-1 . Predictor 52 for each motion model
, The smoothed value vector x ＾ _k-1 (+) based on the n motion models of the equation (58) input from the smoother 62 based on the n motion models and the equation (n) constituting the n motion models From the constant acceleration vector α _{Lk of} (12), the equation (4)
6) Predicted value vector x ＾ _k for each motion model according to
(−) Is calculated, and in the prediction error evaluator 53 for each motion model, the smoothed error covariance matrix P _k−1 (n) based on the n motion models input from the smooth error evaluator 63 based on the n motion models. +) And a preset error covariance matrix P for each motion model according to equation (47) from the driving noise covariance matrix Q _{k-1 of} equation (7) set in advance.
^Lk _k (−) is calculated (step 3).

Next, in a predictor 54 based on n motion models, a smoother 62 based on n motion models obtains the formula (5)
8) Smooth value vector x ＾ _k-1 by n motion models
(+) Is input, a Φ _k-1 vector x ＾ _k-1 (+) is calculated by constant velocity straight-line motion prediction, and each of the equations (32) obtained from the reliability calculator 57 using n motion models is calculated. The reliability β ^Lk-1 _{k-1 of} the motion model, a preset equation (13)
From the transition probability P _LkLk-1 and the constant acceleration vector α _lk of equation (12), the sampling time t is _{calculated according} to equation (90).
The influence term of the acceleration at the time when the detection information in _k is not obtained is estimated, and a predicted value vector x ＾ _k (−) based on n motion models is calculated according to the equation (88) (step 4).

Next, prediction error evaluation using n motion models
Unit 55, a prediction error evaluator 53 for each motion model
From equation (61), the prediction error covariance matrix for each motion model
P^Lk _k(-) Is input, and the reliability of n motion models
Β in the equation (32) obtained from the calculator 57^Lk-1 _k-1,Oh
Transition probability P of equation (13) set in advance_LkLk-1,Oh
Preset constant acceleration vector α_LkAnd equation (9)
0) vector u ＾_{k- 1}More P_kLarge (−) prediction error
Terms to evaluate

[0098]

(Equation 24)

Then, a prediction error covariance matrix P _k (−) based on n motion models is calculated according to equation (93) (step 5). Next, the time is the sampling time t _k−1
After lapse of the sampling time t _k from entering the detection result of the position and range rate from the target and clutter, etc. than the target observation device 51 as detection data sampling time t _k (Step 6).

Next, in the correlator 56, the predicted value vector x ＾ ^Lk _k (−) for each motion model of the equation (56) calculated at the sampling time t _k−1 is passed through the fourth delay circuit 67 to each of the motion models. , The vector z _k (−) in equation (64), which is the center position to be correlated, is replaced with the constant acceleration vector α _Lk in equation (16) instead of the vector x ＾ _k (−) in equation (16). Vector x for zero vector
^算出 Calculate using ^LK _k (-) and calculate the sampling time t
_The prediction error covariance matrix P _k (−) of the n motion models of the equation (66) calculated as _k−1 is input from the prediction error evaluator 55 of the n motion models through the seventh delay circuit 70. Then, through this and the preset observation noise covariance matrix R _k of the equation (19) and the first delay circuit 64, n
S _k of equation (65) used for calculating the size of the area to be correlated is _obtained from the vector x ＾ _k (−) input from the predictor 54 based on the number of motion models, and the tracking target target is calculated by equation (63). And m _k pieces of detection data that may be correlated with each other are selected as Z _k in Expression (20) (Step 7).

Next, in the reliability calculator 57 based on n motion models, the reliability calculator 57 calculates the sampling time t.
to _k-1 type the reliability beta ^{Lk -1} _k-1 of each motion model calculated by the equation (39) through the third delay circuit 66, the motion of the formula (56) calculated for the sampling time t _k-1 The predicted value vector x ＾ ^Lk _k (−) for each model is input from the predictor 52 for each motion model through the fourth delay circuit 67 and is calculated for each motion model of the equation (61) calculated at the sampling time t _k-1. prediction error covariance matrix P ^Lk _k (-) of each motion model through the second delay circuit 65 the prediction error estimator 53
From the observation noise covariance matrix R _{k of} the preset equation (19), and the hypothesis 仮 based on the equations (69), (16), (17) and (53), (54).
The detection data vector z from the correlator 56 under _{k and lk
k,} reliability l _Lk (vector z _{k, i)} of the _i look, preset detection probability P _D, probability P _Gk and wherein the tracking target goal of formula (67) is present in the region where the correlation Equation (7) is obtained from the transition probability P _LkLk-1 of the motion model in (13).
3) to (75) and (39), (41), and (43), the reliability of the motion model and the detection data β _{k, i, Lk, Lk−1} , β ^Lk _k , β _{k, i, Lk} and _{βk, i} are calculated (step 8).

Next, the gain matrix calculator 58
Pulling time t_k-1(61) motion model calculated
Prediction error covariance matrix P^Lk _k(-) Is the second delay time
From the prediction error evaluator 53 for each motion model through the road 65
Input and sampling time t _k-1Equation (57) calculated to
Predicted value vector x ＾ by n motion models_k(-)
Is passed through the first delay circuit 64 and the prediction by n motion models is performed.
Equation (19) input from the measuring instrument 54 and set in advance
The observation noise covariance matrix R_kFrom equation (48) and equation (5)
3) calculating a gain matrix according to equation (54)
Step 9).

Next, in the acceleration vector estimator 59,
Equation (3) is obtained from the reliability calculator 57 using n motion models.
2) β ^Lk _k is input, and the preset equation (1)
Using the constant acceleration vector α _Lk of 2), an acceleration vector estimated value is calculated from Expression (76) according to Expression (78) (Step 10).

Next, in the target motion decision unit 60 based on the estimated acceleration vector, the estimated acceleration vector u _k-1 of the equation (76) is inputted from the acceleration vector estimator 59, and the sampling time t obtained by the estimator 59 is obtained. the desired motion determination result FMD _k-1 type through fifth delay circuit 68 in the _k-1, calculating the smoothed vector estimate u ^S _k-1 of the acceleration vector estimate according to the equation (103) to (105) Then, after calculating the Euclidean norm estimate _{ u ^S _k-1} of the smooth vector estimate u ^S _K-1 of the acceleration vector estimate according to the equation (106), the target motion determination result F
When MD (107) is satisfied, MD _k is calculated by Expression (99).
Is set as follows, and if equation (108) is established, it is set as equation (100), and if equation (109) is established, it is set as equation (110) (step 11). ).

Next, in the smoother 62 based on the n motion models, the smoothed value vector x ＾ _k− based on the n motion models of the equation (58) calculated by the smoother 62 at the sampling time t _k−1 is obtained. ₁ (+) to an input through a delay circuit 69 of the 6, n pieces of reliability beta ^Lk _k and wherein each motion model equations from the reliability calculator 57 according to the motion model (32) (3
The detection data vector z _{k, 1} under the hypothesis Ψ _{k, lk} of 3)
Confidence beta _{k, i,} type the _Lk, correlator 56 the detection data vector z _{k, 1} from the vector z _{k, 2,} ..., vector z _{k, mk}
And the predicted value vector x ＾ for each motion model in equation (56) calculated at the sampling time t _k-1
^Lk _k (−) is input from the predictor 52 for each motion model through the fourth delay circuit 67, the gain matrix K _k of Expression (48) is input from the gain matrix calculator 58, and the acceleration of Expression (76) is obtained. The vector estimation value u _k-1 is input from the acceleration vector estimator 59, the target motion determination result FMD _k is input from the target motion determination device 60 based on the acceleration vector estimation value, and the target motion determination result FMD _k is 1 (curvature). , The expression (8)
1) to (83), calculate a smoothed value vector x ＾ _k (+) based on n motion models. If the target motion determination result FMD _k is 0 (straight), equations (95) and (95) A smoothed value vector x _k (+) based on n motion models is calculated according to the equations (96) and (97).

[0106] In the smoothing error estimator 63 according to the n motion models, beta _{k, i} of formula (31) from the reliability calculator 57 by the n motion models, beta ^Lk _k of the formula (32) ( 33) Input β _{k, i, Lk} of sampling time t
The prediction error covariance matrix P ^Lk _k (−) for each motion model of Equation (61) calculated as _k−1 is input from the prediction error evaluator 53 for each motion model through the second delay circuit 65, and the gain matrix The gain matrix K _k of equation (48) is input from the calculator 58, and the equation (5) calculated at the sampling time t _k−1 is input.
6) Prediction vector x ＾ ^Lk _k (−) for each motion model
Is passed through a fourth delay circuit 67 and the predictor 5 for each motion model
2 and the predicted value vector x ＾ _k (−) based on the n motion models calculated at the sampling time t _k−1 is the first
Predictor 5 based on n motion models through delay circuit 64
4 and the detection data vector z from the correlator 56.
_{k, 1} , vector z _{k, 2} , ..., vector z _{k, mk} ,
The acceleration vector estimate u _k-1 of the formula (76) is input from the acceleration vector estimator 59, the desired motion determination result FMD _k
Is input from the target motion determiner 60 based on the acceleration vector estimation value, and if the target motion determination result FMD _k is 1 (curvature), the equation according to the equations (87), (82), and (83) is used. A smooth error covariance matrix P _k (+) based on the n motion models in (60) is calculated, and the target motion determination result FM
When D _k is 0 (straight), the smooth error covariance matrix P _k (+) based on the n motion models of equation (60) according to equations (98), (96), and (97) Is calculated (step 13). A series of flows in the rue are repeated until the tracking ends (step 14).

Next, a fourth embodiment of the present invention will be described with reference to FIGS.
It is explained according to. Based on the information of the target position and the target distance change rate obtained from the target observation device 51, based on the theory of the ordinary Kalman filter, the initial values of the target position, the smoothed value of the velocity, and the smoothed error covariance matrix are set (step 1). ), N target motion models composed of n constant acceleration vectors of the equation (12) including the zero acceleration vector as in the example of the equation (11) are set (step 2), and tracking is continued. Into a processing loop to perform

Hereinafter, the flow of processing in the loop will be described. For convenience of explanation, the sampling time at the beginning of the loop is t
_Assume that it was _k-1 . Predictor 52 for each motion model
, The smoothed value vector x ＾ _k-1 (+) based on the n motion models of the equation (58) input from the smoother 62 based on the n motion models and the equation (n) constituting the n motion models From the constant acceleration vector α _{Lk of} (12), the equation (4)
6) According to the motion model, the predicted value vector x ＾ _K
Calculate (-).

The prediction error evaluator 53 for each motion model
Smooth error covariance matrix P _k−1 based on n motion models input from smooth error evaluator 63 based on n motion models
A prediction error covariance matrix P ^Lk _k (-) for each motion model is calculated from (+) and a preset driving noise covariance matrix Q _{K-1 of} Expression (7) according to Expression (47) (Step 3). ).

Next, in a predictor 54 based on n motion models, a smoother 62 based on n motion models obtains the equation (5).
8) Smooth value vector x ＾ _k-1 by n motion models
(+) Is input, a Φ _k-1 vector x ＾ _k-1 (+) is calculated by constant velocity straight-line motion prediction, and each of the equations (32) obtained from the reliability calculator 57 using n motion models is calculated. The reliability β ^Lk-1 _{k-1 of} the motion model, a preset equation (13)
From the transition probability P _LkLk−1 of the constant and the constant acceleration vector α _Lk of the equation (12), the sampling time t according to the equation (90).
The influence term of the acceleration at the time when the detection information in _k is not obtained is estimated, and a predicted value vector x ＾ _k (−) based on n motion models is calculated according to the equation (88) (step 4).

Next, prediction error evaluation using n motion models
Unit 55, a prediction error evaluator 53 for each motion model
From equation (61), the prediction error covariance matrix for each motion model
P^Lk _k(-) Is input, and the reliability of n motion models
Β in the equation (32) obtained from the calculator 57^Lk-1 _k-1,Oh
Transition probability P of equation (13) set in advance_Lklk-1,Oh
Preset constant acceleration vector α_LkAnd equation (9)
0) vector u ＾_{k- 1}More P_kLarge (−) prediction error
The term to be evaluated, equation (123) is calculated, and equation (93) is calculated.
Accordingly, the prediction error covariance matrix P based on n motion models
_k(−) Is calculated (step 5).

Next, after the time has elapsed from the sampling time t _k−1 to the sampling time t _k , the correlator 56 uses the target observation device 51 to detect the position and distance change rate from the target and the clutter, etc. at the sampling time t k. Input as _k detection data (step 6). And correlator 5
In Equation 6, the expression (5) calculated at the sampling time t _k-1
6) Prediction vector x ＾ ^Lk _k (−) for each motion model
Is passed through a fourth delay circuit 67 and the predictor 5 for each motion model
2, the central position to be correlated is given by equation (6)
The vector z _k (−) of 4) is calculated using the vector x ＾ ^Lk _k (−) when the constant acceleration vector α _Lk is a zero vector instead of the vector x ＾ _k (−) in equation (16). , A prediction error covariance matrix P based on n motion models of equation (66) calculated at sampling time t _k−1
^Lk _k (−) is input from a prediction error evaluator 55 based on n motion models through a seventh delay circuit 70, and the observation noise covariance matrix R _{k of} equation (19) is set in advance.
And from the predictor 54 based on n motion models through the first delay circuit 64, the vector x ＾ _K (-) is used to calculate the size of the area to be correlated from the vector x ＾ _K (-) (6)
5) S _k is obtained, and m _k pieces of detection data that may be correlated with the tracking target are selected by equation (63), and this is set as Z _k of equation (20) (step 7).

Next, the reliability calculator 57 based on the n motion models calculates the reliability β ^{Lk-1 of} each motion model calculated by the reliability calculator 57 at the sampling time t _{k-1 according} to the equation (39).
_k−1 is input through the third delay circuit 66, and the predicted value vector x ＾ ^Lk _k (−) for each motion model of the equation (56) calculated at the sampling time t _k−1 is input to the fourth delay circuit 67. The prediction error covariance matrix P ^Lk _k (−) for each motion model of Equation (61), which is input from the predictor 52 for each through motion model and calculated at the sampling time t _k−1 , is calculated by the second delay circuit 6
5, input from the prediction error evaluator 53 for each motion model, and from the preset observation noise covariance matrix R _{k of} the equation (19), the equations (69), (16), and (17) are used.
And the reliability l _Lk (vector z _{k, i} ) of the detection data vector z _{k, i} from the correlator 56 under the hypothesis Ψ _{k, lk} is _{obtained from the} equations (53) and (54) and set in advance. The equations (73) to (73) are obtained by using the detected detection probability P _D , the probability P _Gk that the tracking target of the equation (67) exists in the area to be correlated, and the transition probability P _{Lklk −1} of the motion model of the equation (13). (7
5) and the reliability β of the motion model and the detection data according to the equations (39), (41), and (43).
_{k, i, Lk, Lk-1} , β ^Lk _k , β _{k, i, Lk} , and β _{k, i} are calculated (step 8).

Next, the gain matrix calculator 58
Pulling time t_k-1(61) motion model calculated
Prediction error covariance matrix P^Lk _k(-) Is the second delay time
From the prediction error evaluator 53 for each motion model through the road 65
Input and sampling time t _k-1Equation (57) calculated to
Predicted value vector x ＾ by n motion models_k(-)
Is passed through the first delay circuit 64 and the prediction by n motion models is performed.
Equation (19) input from the measuring instrument 54 and set in advance
The observation noise covariance matrix R_kFrom equation (48) and equation (5)
3) calculating a gain matrix according to equation (54)
Step 9).

Next, in the acceleration vector estimator 59,
Equation (3) is obtained from the reliability calculator 57 using n motion models.
2) Enter β ^LK _K and enter the preset equation (1)
Using the constant acceleration vector α _Lk of 2), an acceleration vector estimated value of Expression (76) is calculated according to Expression (78) (Step 10).

Next, the target operation based on the predicted value of the target distance change rate is performed.
In the motion determiner 61, the sampling time t_k-1Calculated
Predicted value vector by n motion models of equation (57)
X_k(-) Is passed through the first delay circuit 64 to make n motions.
Input from the predictor 54 based on the model,
Degree β_{k, i}From the reliability calculator 57 using n motion models
And the sampling time t in the target motion determiner 61
_k-1Motion judgment result FMD_k-1The fifth delay
Input through the circuit 68 and the detection data of the detection data from the correlator 56.
Formula used to calculate the size of the area to be correlated
(65) S_kAnd furthermore, the detection data is output from the correlator 56.
Data vector z_{k, 1}, Vector z_{k, 2}, ..., vector z
_{k, mk}And input the distance change rate component to R^o _{k, 1}, R
^o _{k, 2}, ..., R^o _{k, mk}Equation (113), Equation (1)
7) the predicted value R of the target distance change rate ^{. P} _K, Equation (11)
4) S_k4th row and 4th column element S_k(4,4), equation
According to (116), the weighted average estimation value R of the target distance change rate is obtained.
^o _kAnd further according to equation (118).
Scalar quantity ε_kAfter calculating the target motion determination result FMD_k
When equation (119) is satisfied, as in equation (99),
And if equation (120) holds, then equation (1)
00), and when equation (121) is satisfied.
In this case, it is set as shown in equation (110) (step 1).
2).

Next, in the smoother 62 based on the n motion models, the smoothed value vector x ＾ based on the n motion models of the equation (58) calculated at the sampling time t _k−1 is obtained.
_k−1 (+) is input through the sixth delay circuit 69, and the reliability calculator 57 based on n motion models calculates the reliability β ^Lk _k of each motion model of Expression (32) and the reliability β ^Lk _k of Expression (33). Hypothesis Ψ _k,
the reliability β of the detection data vector z _{k, i} under _{lk
k, i, and Lk} are input, and the detection data vector z is output from the correlator 56.
_{k, 1} , vector z _{k, 2} , ..., vector z _{k, mk} ,
The predicted value vector x ＾ ^Lk _k (−) for each motion model of equation (56) calculated at the sampling time t _k−1 is calculated as the fourth
Of the motion model for each motion model, the gain matrix K _k of equation (48) from the gain matrix calculator 58, and the acceleration vector estimated value u _k-1 of equation (76). It is input from the acceleration vector estimator 59 and the target motion determination result FMD _k is input from the target motion determiner 61 based on the predicted value of the target distance change rate.
_{When k} is 1 (curvature), Expressions (81) to (8)
According to 3), a smoothed value vector x _k (+) based on n motion models is calculated, and the target motion determination result FMD _k is 0.
(Straight ahead), Equation (95) and Equation (9)
6) According to equation (97), a smoothed value vector x） _k (+) based on n motion models is calculated.

In the smoothing error estimator 63 based on the n motion models, the reliability calculator 57 based on the n motion models calculates β _{k, i} in equation (31), β ^Lk _{k in} equation (32) and equation (32). 33) Input β _{k, i, Lk} of sampling time t
The prediction error covariance matrix P ^Lk _k (−) for each motion model of Equation (61) calculated as _k−1 is input from the prediction error evaluator 53 for each motion model through the second delay circuit 65, and the gain matrix The gain matrix K _k of equation (48) is input from the calculator 58, and the equation (5) calculated at the sampling time t _k−1 is input.
6) Prediction vector x ＾ ^Lk _k (−) for each motion model
Is passed through a fourth delay circuit 67 and the predictor 5 for each motion model
2 and the predicted value vector x ＾ _k (−) based on the n motion models calculated at the sampling time t _k−1 is the first
Predictor 5 based on n motion models through delay circuit 64
4 and the detection data vector z from the correlator 56.
_{k, 1} , vector z _{k, 2} , ..., vector z _{k, mk} ,
The acceleration vector estimate u _k-1 of the formula (76) is input from the acceleration vector estimator 59, the desired motion determination result FMD _k
Was input from the desired motion determiner 61 according to the predicted value of the target range rate, if desired motion determination result FMD _k is 1 (curvilinear progression) have the formula (87) and (82), formula (83)
Calculates a smooth error covariance matrix P _k (+) based on the n motion models of the equation (60) according to the equation (60), and calculates the target motion determination result F
If MD _k is 0 (straight), the smooth error covariance matrix P _k (+) based on the n motion models of equation (60) according to equations (98), (96), and (97) Is calculated (step 13). This series of flows in the loop is repeated until tracking ends (step 14).

Next, a fifth embodiment of the present invention will be described with reference to FIGS.
It is explained according to. Based on the information of the target position and the target distance change rate obtained from the target observation device 51, based on the theory of the ordinary Kalman filter, the initial values of the target position, the smoothed value of the velocity, and the smoothed error covariance matrix are set (step 1). ), N target motion models composed of n constant acceleration vectors of the equation (12) including the zero acceleration vector as in the example of the equation (11) are set (step 2), and tracking is continued. Into a processing loop to perform

Hereinafter, the flow of processing in the loop will be described. For convenience of explanation, the sampling time at the beginning of the loop is t
_Assume that it was _k-1 . Predictor 52 for each motion model
, The smoothed value vector x ＾ _k-1 (+) based on the n motion models of the equation (58) input from the smoother 62 based on the n motion models and the equation (n) constituting the n motion models From the constant acceleration vector α _{Lk of} (12), the equation (4)
6) Predicted vector x 予 測 for each motion model according to
Calculate _K (-).

A prediction error evaluator 53 for each motion model
, A smoothing error estimator 63 based on n motion models
The smoothed error covariance matrix P _k−1 (+) based on n motion models input from
Then, a prediction error covariance matrix P ^Lk _k (−) for each motion model is calculated from the driving noise covariance matrix Q _{k−1 according} to equation (47) (step 3).

Next, in a predictor 54 based on n motion models, a smoother 62 based on n motion models obtains the formula (5)
8) Smooth value vector x ＾ _k-1 by n motion models
(+) Is input, a Φ _k-1 vector x ＾ _k-1 (+) is calculated by constant velocity straight-line motion prediction, and each of the equations (32) obtained from the reliability calculator 57 using n motion models is calculated. The reliability β ^Lk-1 _{k-1 of} the motion model, a preset equation (13)
From the transition probability P _LkLk−1 of the constant and the constant acceleration vector α _Lk of the equation (12), the sampling time t according to the equation (90).
The influence term of the acceleration at the time when the detection information in _k is not obtained is estimated, and a predicted value vector x ＾ _k (−) based on n motion models is calculated according to the equation (88) (step 4).

Next, prediction error evaluation using n motion models
Unit 55, a prediction error evaluator 53 for each motion model
From equation (61), the prediction error covariance matrix for each motion model
P^Lk _k(-) Is input, and the reliability of n motion models
Β in the equation (32) obtained from the calculator 57^Lk-1 _k-1,Oh
Transition probability P of equation (13) set in advance_LkLk-1,Oh
Preset constant acceleration vector α_LkAnd equation (9)
0) vector u ＾_{K- 1}More P_kLarge (−) prediction error
The term to be evaluated, equation (123) is calculated, and equation (93) is calculated.
Accordingly, the prediction error covariance matrix P based on n motion models
_k(−) Is calculated (step 5).

Next, after the time has elapsed from the sampling time t _k−1 to the sampling time t _k , the correlator 56 uses the target observing device 51 to detect the position and distance change rate from the target and the clutter, etc. at the sampling time t k. Input as _k detection data (step 6). Next, the correlator 56
In equation (5) calculated at sampling time t _k−1
7) Predicted value vector x ＾ by n motion models
_k (−) is input from the predictor 54 based on n motion models through the first delay circuit 64, and the vector z _k (−) of the equation (64) which is the center position to be correlated is calculated by the equation (16). )
And the prediction error covariance matrix P _k by the n models of equation (61) calculated at sampling time t _k−1
(−) Is input from a prediction error evaluator 55 based on n motion models through a seventh delay circuit 70, and is input to the observation noise covariance matrix R _k and the first observation noise covariance matrix R _{k in} Expression (19). From the vector x ＾ _k (−) input from the predictor 54 based on the n motion models through the delay circuit 64, S _k of the equation (65) used for calculating the size of the area to be correlated is obtained, and the equation (63) is obtained. ), M _k pieces of detection data which may be correlated with the tracking target are selected, and this is calculated by the equation (2).
0) is set as z _k (step 7).

Next, in the reliability calculator 57 based on n motion models, the reliability calculator 57 samples the sampling time t.
to _k-1 type the reliability beta ^{Lk -1} _k-1 of each motion model calculated by the equation (39) through the third delay circuit 66, the motion of the formula (56) calculated for the sampling time t _k-1 The predicted value vector x ＾ ^Lk _k (−) for each model is input from the predictor 52 for each motion model through the fourth delay circuit 67 and is calculated for each motion model of the equation (61) calculated at the sampling time t _k-1. prediction error covariance matrix P ^Lk _k (-) of each motion model through the second delay circuit 65 the prediction error estimator 53
From the observation noise covariance matrix R _{k of} the preset equation (19), and the hypothesis 仮 based on the equations (69), (16), (17) and (53), (54).
The detection data vector z from the correlator 56 under _{k and lk
k,} reliability l _Lk (vector z _{k, i)} of the _i look, preset detection probability P _D, probability P _Gk and wherein the tracking target goal of formula (67) is present in the region where the correlation Equation (7) is obtained from the transition probability P _LkLk-1 of the motion model in (13).
3) to (75) and (39), (41), and (43), the reliability of the motion model and the detection data β _{k, i, Lk, Lk−1} , β ^Lk _k , β _{k, i, Lk} and _{βk, i} are calculated (step 8).

Next, the gain matrix calculator 58
Pulling time t_k-1(61) motion model calculated
Prediction error covariance matrix P^Lk _k(-) Is the second delay time
From the prediction error evaluator 53 for each motion model through the road 65
Input and sampling time t _k-1Equation (57) calculated to
Predicted value vector x ＾ by n motion models_k(-)
Is passed through the first delay circuit 64 and the prediction by n motion models is performed.
Equation (19) input from the measuring instrument 54 and set in advance
The observation noise covariance matrix R_kFrom equation (48) and equation (5)
3) calculating a gain matrix according to equation (54)
Step 9).

Next, in the acceleration vector estimator 59,
Equation (3) is obtained from the reliability calculator 57 using n motion models.
2) β ^Lk _k is input, and the preset equation (1)
Using the constant acceleration vector α _Lk of 2), an acceleration vector estimated value of Expression (76) is calculated according to Expression (78) (Step 10).

Next, in the target motion decision unit 60 based on the acceleration vector estimation value, the acceleration vector estimation value u _k-1 of the equation (76) is input from the acceleration vector estimation unit 59, and the sampling time t _k− the desired motion determination result FMD _k-1 at ₁ and inputted through the fifth delay circuit 68, calculates a smoothed vector estimate u ^S _K-1 of the acceleration vector estimate according to the equation (103) to (105), Further, after calculating the Euclidean norm vector estimated value _{ u ^S _k-1} of the smoothed vector estimated value u ^S _K-1 of the acceleration vector estimated value according to the equation (106), the target motion determination result FMD _k is calculated by the equation (107). ) Is satisfied, the expression (9) is satisfied.
9), if equation (108) holds, set as equation (100), and if equation (109) holds, set as equation (110) ( Step 11).

Next, a smoother 62 based on n motion models
Here, the sampling time t_k-1Calculated
The smoothed value by the n motion models of equation (58)
Kuttle x ＾_k-1(+) Is input through the sixth delay circuit 69
Then, a reliability calculator 57 based on n motion models calculates the equation
(32) Reliability β of each motion model^Lk _kAnd equation (3
Hypothesis 3)_{k, lk}Data vector z under the condition_{k, i}
Reliability β_{k, i, Lk}Input from the correlator 56 and
Vector z_{k, 1}, Vector z_{k, 2}, ..., vector z _{k, mk}
At the sampling time t_k-1Formula calculated in
Prediction vector x ＾ for each motion model in (56)
^Lk _k(-) Is passed through the fourth delay circuit 67 for each motion model.
From the predictor 52, and the gain matrix calculator 58 calculates
(48) gain matrix K_kAnd the acceleration of equation (76)
Degree vector estimate u_k-1From the acceleration vector estimator 59
And input the target motion judgment result FMD_kThe acceleration vector
Input from the target motion determiner 60 based on the estimated value
Judgment result FMD_kIs 1 (in progress), the expression (8)
1) According to the equation (83), n
Smooth value vector x ＾_kCalculate (+) and determine the target motion
FMD_kIs 0 (straight ahead), the equation (95) and
And n motion models according to equations (96) and (97)
Vector x_kCalculate (+).

[0130] In the smoothing error estimator 63 according to the n motion models, beta _{k, i} of formula (31) from the reliability calculator 57 by the n motion models, beta ^Lk _k of the formula (32) ( 33) Input β _{k, i, Lk} of sampling time t
The prediction error covariance matrix P ^Lk _k (−) for each motion model of Equation (61) calculated as _k−1 is input from the prediction error evaluator 53 for each motion model through the second delay circuit 65, and the gain matrix The gain matrix K _k of equation (48) is input from the calculator 58, and the equation (5) calculated at the sampling time t _k−1 is input.
6) Prediction vector x ＾ ^Lk _k (−) for each motion model
Is passed through a fourth delay circuit 67 and the predictor 5 for each motion model
2 and the predicted value vector x ＾ _k (−) based on the n motion models calculated at the sampling time t _k−1 is the first
Predictor 5 based on n motion models through delay circuit 64
4 and the detection data vector z from the correlator 56.
_{k, 1} , vector z _{k, 2} , ..., vector z _{k, mk} ,
The acceleration vector estimate u _k-1 of the formula (76) is input from the acceleration vector estimator 59, the desired motion determination result FMD _k
Is input from the target motion determiner 60 based on the acceleration vector estimation value, and if the target motion determination result FMD _k is 1 (curvature), the equation according to the equations (87), (82), and (83) is used. A smooth error covariance matrix P _k (+) based on the n motion models in (60) is calculated, and the target motion determination result FM
When D _k is 0 (straight), the smooth error covariance matrix P _k (+) based on the n motion models of equation (60) according to equations (98), (96), and (97) Is calculated (step 13). This series of flows in the loop is repeated until tracking ends (step 14).

Next, Embodiment 6 of the present invention will be described with reference to FIGS.
0 will be described. Based on the information of the target position and the target distance change rate obtained from the target observation device 51, based on the theory of the ordinary Kalman filter, the initial values of the target position, the smoothed value of the velocity, and the smoothed error covariance matrix are set (step 1). ), N target motion models composed of n constant acceleration vectors of the equation (12) including the zero acceleration vector as in the example of the equation (11) are set (step 2), and tracking is continued. Into a processing loop to perform

Hereinafter, the flow of processing in the loop will be described. For convenience of explanation, the sampling time at the beginning of the loop is t
_Assume that it was _k-1 . Predictor 52 for each motion model
, The smoothed value vector x ＾ _k-1 (+) based on the n motion models of the equation (58) input from the smoother 62 based on the n motion models and the equation (n) constituting the n motion models From the constant acceleration vector α _{Lk of} (12), the equation (4)
6) Predicted value vector x ＾ _k for each motion model according to
(−) Is calculated, and in the prediction error evaluator 53 for each motion model, the smoothed error covariance matrix P _k−1 (n) based on the n motion models input from the smooth error evaluator 63 based on the n motion models. +) And a preset error covariance matrix P for each motion model according to equation (47) from the driving noise covariance matrix Q _{k-1 of} equation (7) set in advance.
^Lk _k (−) is calculated (step 3).

Next, in a predictor 54 based on n motion models, a smoother 62 based on n motion models obtains the equation (5).
8) Smooth value vector x ＾ _k-1 by n motion models
(+) Is input, a Φ _k-1 vector x ＾ _k-1 (+) is calculated by constant velocity straight-line motion prediction, and each of the equations (32) obtained from the reliability calculator 57 using n motion models is calculated. The reliability β ^Lk-1 _{k-1 of} the motion model, a preset equation (13)
From the transition probability P _LkLk−1 of the constant and the constant acceleration vector α _Lk of the equation (12), the sampling time t according to the equation (90).
The influence term of the acceleration at the time when the detection information in _k is not obtained is estimated, and a predicted value vector x ＾ _k (−) based on n motion models is calculated according to the equation (88) (step 4).

Next, prediction error evaluation using n motion models
Unit 55, a prediction error evaluator 53 for each motion model
From equation (61), the prediction error covariance matrix for each motion model
P^Lk _k(-) Is input, and the reliability of n motion models
Β in the equation (32) obtained from the calculator 57^Lk-1 _k-1,Oh
Transition probability P of equation (13) set in advance_LkLk-1,Oh
Preset constant acceleration vector α_LkAnd equation (9)
0) vector estimate u_k-1More P_k(-) Prediction error
Equation (123), which is a term that greatly evaluates
3) Prediction error covariance with n motion models according to
Matrix P_k(−) Is calculated (step 5).

Next, after the time has elapsed from the sampling time t _{k −1} to the sampling time t _k , the target observation device 51
To enter more positions and range rate from the target and clutter like detection result as detection data sampling time t _k (Step 6).

Next, in the correlator 56, the prediction value vector x ＾ _k (−) based on the n motion models of the equation (57) calculated at the sampling time t _k−1 is passed through the first delay circuit 64 to the n Input from the predictor 54 based on the motion model of
The vector z in equation (64) that is the center position to be correlated
_k (−) is calculated by equation (16), and the prediction error covariance matrix P ^Lk _k (−) for each motion model of equation (61) calculated at sampling time t _k−1 is calculated by the second delay circuit 65. A prediction error estimator 53 is input from the prediction error evaluator 53 for each of the through motion models, passes through it and a preset observation noise covariance matrix R _{k of} equation (19) and a first delay circuit 64, and predictors 54 based on n motion models. Vector x ＾ input from
_{From S k} (−), S _k of Expression (65) used for calculating the size of the area to be correlated is obtained, and m _K detection data possibly correlated with the tracking target is calculated by Expression (63). And select it as Z _k in equation (20) (step 7).

Next, in the reliability calculator 57 based on the n motion models, the reliability calculator 57 samples the sampling time t.
to _k-1 type the reliability beta ^{Lk -1} _k-1 of each motion model calculated by the equation (39) through the third delay circuit 66, the motion of the formula (56) calculated for the sampling time t _k-1 The predicted value vector x ＾ ^Lk _k (−) for each model is input from the predictor 52 for each motion model through the fourth delay circuit 67 and is calculated for each motion model of the equation (61) calculated at the sampling time t _k-1. prediction error covariance matrix P ^Lk _k (-) of each motion model through the second delay circuit 65 the prediction error estimator 53
From the observation noise covariance matrix R _{k of} the preset equation (19), and the hypothesis 仮 based on the equations (69), (16), (17) and (53), (54).
The detection data vector z from the correlator 56 under _{k and lk
k,} the reliability 1 _Lk (vector z _{k, i)} of the _i look, preset detection probability P _D, probability P _Gk and wherein the tracking target goal of formula (67) is present in the region where the correlation Equation (7) is obtained from the transition probability P _LkLk-1 of the motion model in (13).
3) to (75) and (39), (41), and (43), the reliability of the motion model and the detection data β _{k, i, Lk, Lk−1} , β ^Lk _k , β _{k, i, Lk} and _{βk, i} are calculated (step 8).

Next, the gain matrix calculator 58
Pulling time t_k-1(61) motion model calculated
Prediction error covariance matrix P^Lk _k(-) Is the second delay time
From the prediction error evaluator 53 for each motion model through the road 65
Input and sampling time t _k-1Equation (57) calculated to
Predicted value vector x ＾ by n motion models_k(-)
Is passed through the first delay circuit 64 and the prediction by n motion models is performed.
Equation (19) input from the measuring instrument 54 and set in advance
The observation noise covariance matrix R_kFrom equation (48) and equation (5)
3) calculating a gain matrix according to equation (54)
Step 9).

Next, in the acceleration vector estimator 59,
Equation (3) is obtained from the reliability calculator 57 using n motion models.
2) β ^Lk _k is input, and the preset equation (1)
Using the constant acceleration vector α _Lk of 2), an acceleration vector estimated value of Expression (76) is calculated according to Expression (78) (Step 10).

Next, the target operation based on the predicted value of the target distance change rate is performed.
In the motion determiner 61, the sampling time t_k-1Calculated
Predicted value vector by n motion models of equation (57)
X_k(-) Is passed through the first delay circuit 64 to make n motions.
Input from the predictor 54 based on the model,
Degree β_{k, i}From the reliability calculator 57 using n motion models
And the sampling time t in the target motion determiner 61
_k-1Motion judgment result FMD_k-1The fifth delay
Input through the circuit 68 and the detection data of the detection data from the correlator 56.
Formula used to calculate the size of the area to be correlated
(65) S_kAnd furthermore, the detection data is output from the correlator 56.
Data vector z_{k, 1}, Vector z_{k, 2}, ..., vector z
_{k, mk}And input the distance change rate component to R^o _{k, 1}, R
^o _{k, 2}, ..., R^o _{k, mk}Equation (113), Equation (1)
7) the predicted value R of the target distance change rate ^{. P} _k, Equation (11)
4) S_k4th row and 4th column element S_k(4,4), equation
According to (116), the weighted average estimation value R of the target distance change rate is obtained.
^o _kAnd further according to equation (118).
Scalar quantity ε_kAfter calculating the target motion determination result FMD_k
When equation (119) is satisfied, as in equation (99),
And if equation (120) holds, then equation (1)
00), and when equation (121) is satisfied.
In this case, it is set as shown in equation (110) (step 1).
1).

Next, in the smoother 62 based on the n motion models, the smoothed value vector x ＾ based on the n motion models of the equation (58) calculated at the sampling time t _k−1 is obtained.
_k−1 (+) is input through the sixth delay circuit 69, and the reliability calculator 57 based on n motion models calculates the reliability β ^Lk _k of each motion model of Expression (32) and the reliability β ^Lk _k of Expression (33). Hypothesis Ψ _k,
the reliability β of the detection data vector z _{k, i} under _{lk
k, i, and Lk} are input, and the detection data vector z is output from the correlator 56.
_{k, 1} , vector z _{k, 2} , ..., vector z _{k, mk} ,
The predicted value vector x ＾ ^Lk _k (−) for each motion model of equation (56) calculated at the sampling time t _k−1 is calculated as the fourth
Of the motion model for each motion model, the gain matrix K _k of equation (48) from the gain matrix calculator 58, and the acceleration vector estimated value u _k-1 of equation (76). It is input from the acceleration vector estimator 59 and the target motion determination result FMD _k is input from the target motion determiner 61 based on the predicted value of the target distance change rate.
_{When k} is 1 (curvature), Expressions (81) to (8)
According to 3), a smoothed value vector x _k (+) based on n motion models is calculated, and the target motion determination result FMD _k is 0.
(Straight ahead), Equation (95) and Equation (9)
6) According to equation (97), a smoothed value vector x） _k (+) based on n motion models is calculated.

In the smoothing error evaluator 63 based on n motion models, the reliability calculator 57 based on n motion models calculates β _{k, i} in equation (31), β ^Lk _{k in} equation (32) and equation (32). 33) Input β _{k, i, Lk} of sampling time t
The prediction error covariance matrix P ^Lk _k (−) for each motion model of Equation (61) calculated as _k−1 is input from the prediction error evaluator 53 for each motion model through the second delay circuit 65, and the gain matrix The gain matrix K _k of equation (48) is input from the calculator 58, and the equation (5) calculated at the sampling time t _k−1 is input.
6) Prediction vector x ＾ ^Lk _k (−) for each motion model
Is passed through a fourth delay circuit 67 and the predictor 5 for each motion model
2 and the predicted value vector x ＾ _k (−) based on the n motion models calculated at the sampling time t _k−1 is the first
Predictor 5 based on n motion models through delay circuit 64
4 and the detection data vector z from the correlator 56.
_{k, 1} , vector z _{k, 2} , ..., vector z _{k, mk} ,
The acceleration vector estimate u _k-1 of the formula (76) is input from the acceleration vector estimator 59, the desired motion determination result FMD _k
Was input from the desired motion determiner 61 according to the predicted value of the target range rate, if desired motion determination result FMD _k is 1 (curvilinear progression) have the formula (87) and (82), formula (83)
Calculates a smooth error covariance matrix P _k (+) based on the n number of motion models of the equation (60), and calculates a target motion determination result FM.
When D _k is 0 (straight), the smooth error covariance matrix P _k (+) based on the n motion models of equation (60) according to equations (98), (96), and (97) Is calculated (step 13). This series of flows in the loop is repeated until tracking ends (step 14).

[0143]

As described above, according to the present invention, it is possible to improve the accuracy of calculating the target motion parameters without adding a special additional device to the normal automatic target tracking device.
Although the above description has been given of the case where the constant acceleration vector is added to the constant velocity linear motion model, the target motion parameters are calculated from information of the target observation device having a plurality of other motion models. The present invention can be applied to a multi-target tracking method and a multi-target tracking device.

[Brief description of the drawings]

FIG. 1 is a diagram showing a processing procedure of a first embodiment of a multi-target tracking apparatus of the present invention.

FIG. 2 is a diagram illustrating a configuration of a first embodiment of the multi-target tracking apparatus according to the present invention;

FIG. 3 is a diagram showing a processing procedure of a multi-target tracking apparatus according to a second embodiment of the present invention.

FIG. 4 is a diagram illustrating a configuration of a multi-target tracking apparatus according to a second embodiment of the present invention;

FIG. 5 is a diagram showing a processing procedure of embodiments 3 and 5 of the multi-target tracking apparatus of the present invention.

FIG. 6 is a diagram showing a configuration of a third embodiment of the multi-target tracking apparatus of the present invention.

FIG. 7 is a diagram showing a processing procedure of embodiments 4 and 6 of the multi-target tracking apparatus of the present invention.

FIG. 8 is a diagram illustrating a configuration of a fourth embodiment of the multi-target tracking apparatus according to the present invention;

FIG. 9 is a diagram showing a configuration of a fifth embodiment of the multi-target tracking apparatus of the present invention.

FIG. 10 is a diagram showing a configuration of a sixth embodiment of the multi-target tracking apparatus of the present invention.

FIG. 11 is a diagram illustrating a constant acceleration vector.

FIG. 12 is a diagram illustrating a correlation between a tracking target and detection data.

FIG. 13 is a diagram showing a processing procedure of one embodiment of a conventional multi-target tracking method.

FIG. 14 is a diagram showing a configuration of an example of a conventional multi-target tracking device.

[Explanation of symbols]

Reference Signs List 1 Initial value setting step 2 n motion model setting steps 3 Prediction value and prediction error covariance matrix calculation step for each motion model 4 Prediction value calculation step using n motion models 5 Prediction error covariance using n motion models Matrix calculation step 6 detection data input step 7 detection data selection step 8 motion model and detection data reliability calculation step 9 gain matrix calculation step 10 acceleration vector estimation step 11 target motion determination step based on acceleration vector estimation value 12 target distance change rate Target motion determination step based on predicted value 13 Smooth value and smooth error covariance matrix calculation step based on n motion models 14 End determination step 15 Constant velocity linear motion model setting step 16 Predicted value calculation step using constant velocity linear motion model 17 Constant velocity Linear motion Prediction error covariance matrix calculation step 18 based on detection data 18 detection data reliability calculation step 19 smoothed value and smooth error covariance matrix calculation step using constant velocity linear motion model 51 target observation device 52 predictor for each motion model 53 for each motion model Prediction error evaluator 54 Predictor with n motion models 55 Prediction error evaluator with n motion models 56 Correlator 57 Reliability calculator with n motion models 58 Gain matrix calculator 59 Acceleration vector estimator 60 Acceleration Target motion determiner based on vector estimation value 61 Target motion determiner based on predicted value of target distance change rate 62 Smoother based on n motion models 63 Smooth error evaluator based on n motion models 64 First delay circuit 65 Second Delay circuit 66 third delay circuit 67 fourth delay circuit 68 fifth delay circuit 69 Sixth delay circuit 70 Seventh delay circuit 81 Predictor based on constant velocity linear motion model 82 Detector data reliability calculator 83 Smoother using constant velocity linear motion model 84 Smooth error evaluator based on constant velocity linear motion model 85 etc. Prediction error estimator based on fast-moving linear motion model 86 Eighth delay circuit 87 Ninth delay circuit

Claims (6)

[Claims] 1. A means for setting an initial value of a smoothed error covariance matrix estimating a smoothed value such as a target position and a speed and an error of the smoothed value, and a target by using a plurality of n constant vectors of the same dimension. means for setting n motion models, n
Means for calculating a predicted value such as a target position and a velocity for each of the motion models and a prediction error covariance matrix estimating the error thereof; means for calculating a predicted value such as a target position and a speed for the entire n number of motion models Means for generally inputting a plurality of observation values of a target position and a target distance change rate as detection data, including false detection data, a prediction value based on the entire n motion models and a prediction error for each of the n motion models. Means for selecting only a few pieces of detection data that may be correlated with the tracking target from among the plurality of pieces of detection data using a variance matrix, a plurality of n motion models and a selected one of the selected detection data. Means for calculating the reliability, means for calculating the same gain matrix for each of the motion models, means for calculating the estimated values of the constant vectors constituting the n motion models, and the constant vector Means for determining the target motion based on the estimated value of the motion vector, and selecting the setting method of the n motion models from a plurality of cases based on the judgment result of the target motion while selecting the target position, speed, etc. of the entire n motion models. Means for calculating a smoothed value and a smoothed error covariance matrix estimating an error of the smoothed value. 2. A means for setting an initial value of a smoothed error covariance matrix estimating a smoothed value such as a target position and a speed and an error of the smoothed value, and a target by using a plurality of n constant vectors of the same dimension. means for setting n motion models, n
Means for calculating a predicted value such as a target position and a velocity for each of the motion models and a prediction error covariance matrix estimating the error thereof; means for calculating a predicted value such as a target position and a speed for the entire n number of motion models Means for generally inputting a plurality of observation values of a target position and a target distance change rate as detection data, including false detection data, a prediction value based on the entire n motion models and a prediction error for each of the n motion models. Means for selecting only a few pieces of detection data that may be correlated with the tracking target from among the plurality of pieces of detection data using a variance matrix, a plurality of n motion models and a selected one of the selected detection data. Means for calculating reliability, means for calculating the same gain matrix for each of the motion models, means for calculating an estimated value of a constant vector constituting the n motion models, target distance change Means for calculating a predicted value of the target motion, and determining a target motion based on the value. A target position based on the entire n motion models while selecting a method of setting the n motion models from a plurality of cases based on the result of the determination of the target motion. Means for calculating a smoothed value such as a speed and a smoothed error covariance matrix estimating an error thereof. 3. A means for setting an initial value of a smoothed error covariance matrix estimating a smoothed value such as a target position and a speed and an error of the smoothed value, and a target by using a plurality of n constant vectors of the same dimension. means for setting n motion models, n
Means for calculating a prediction error covariance matrix estimating a predicted value of a target position, a speed, and the like for each of the motion models, and a prediction value of a target position, a speed, and the like by using the entire n motion models. Means for calculating, a means for calculating a prediction error covariance matrix estimating an error of a prediction value of the entire n motion models, and a plurality of observation values of target position and target distance change rate as detection data including false detection data. Means for inputting, a prediction value for each of the n motion models, and a prediction error covariance matrix based on the n motion models as a whole. Means for selecting only some pieces of detection data having a certain number, means for calculating the reliability of each of the plurality of n motion models and the selected detection data, and the same gain row for each of the motion models. Means for calculating an estimated value of a constant vector constituting the n motion models, means for determining a target motion based on the estimated value of the constant vector, and n means for determining a target motion based on the determination result of the target motion. Means for calculating a smoothed value such as a target position, a velocity, etc., of the entire n number of motion models and a smoothed error covariance matrix estimating an error thereof, while selecting a method of setting the motion model from a plurality of cases. Multi-target tracking device. 4. A means for setting an initial value of a smoothed error covariance matrix in which a smoothed value such as a target position and a speed and an error of the smoothed value are estimated, and a target by using a plurality of n constant vectors of the same dimension. means for setting n motion models, n
Means for calculating a prediction error covariance matrix estimating a predicted value of a target position, a speed, and the like for each of the motion models, and a prediction value of a target position, a speed, and the like by using the entire n motion models. Means for calculating, a means for calculating a prediction error covariance matrix estimating an error of a prediction value of the entire n motion models, and a plurality of observation values of target position and target distance change rate as detection data including false detection data. Means for inputting, a prediction value for each of the n motion models, and a prediction error covariance matrix based on the n motion models as a whole. Means for selecting only some pieces of detection data having a certain number, means for calculating the reliability of each of the plurality of n motion models and the selected detection data, and the same gain row for each of the motion models. Means for calculating an estimated value of a constant vector forming the n number of motion models; means for calculating a predicted value of a target distance change rate and determining a target motion based on the value; and determining the target motion. Means for calculating a smoothed error covariance matrix estimating a smoothed value such as a target position, a velocity, etc. of the entire n number of motion models and an error thereof while selecting a method of setting n number of motion models from a plurality of cases based on the result; A multi-target tracking device, comprising: 5. A means for setting an initial value of a smoothed error covariance matrix estimating a smoothed value such as a target position and a speed and an error of the smoothed value, and a target by using a plurality of n constant vectors of the same dimension. means for setting n motion models, n
Means for calculating a prediction error covariance matrix estimating a predicted value of a target position, a speed, and the like for each of the motion models, and a prediction value of a target position, a speed, and the like by using the entire n motion models. Means for calculating, a means for calculating a preliminary error covariance matrix estimating an error of a prediction value by the entire n motion models, and a plurality of observation data of target position and target distance change rate as detection data including false detection data. Using the input means, the predicted value of the entire n motion models and the prediction error covariance matrix of the entire n motion models, the possibility of correlation with the tracking target from among the plurality of pieces of detection data. Means for selecting only certain detection data, means for calculating the reliability of each of the plurality of n motion models and the selected detection data, the same gain matrix for each of the motion models Means for calculating, means for calculating the estimated value of the constant vector constituting the n-number of motion models,
Means for determining a target motion based on the estimated value of the constant vector, a target position and a speed based on the entire n motion models while selecting a plurality of methods for setting the n motion models based on the determination result of the target motion Means for calculating a smoothed error covariance matrix estimating a smoothed value and an error thereof. 6. A means for setting an initial value of a smoothed error covariance matrix estimating a smoothed value such as a target position and a speed and an error of the smoothed value, and a target by using a plurality of n constant vectors of the same dimension. means for setting n motion models, n
Means for calculating a prediction error covariance matrix estimating a predicted value of a target position, a speed, and the like for each of the motion models, and a prediction value of a target position, a speed, and the like by using the entire n motion models. Means for calculating, a means for calculating a prediction error covariance matrix estimating an error of a prediction value of the entire n motion models, and a plurality of observation values of target position and target distance change rate as detection data including false detection data. Using the input means, the predicted value of the entire n motion models and the prediction error covariance matrix of the entire n motion models, the possibility of correlation with the tracking target from among the plurality of pieces of detection data. Means for selecting only certain detection data, means for calculating the reliability of each of the plurality of n motion models and the selected detection data, the same gain matrix for each of the motion models Means for calculating, means for calculating the estimated value of the constant vector constituting the n-number of motion models,
Means for calculating a predicted value of the target distance change rate and determining a target motion based on the calculated value;
Means for calculating a smoothed value such as a target position, a speed, etc. of the entire n number of motion models and a smoothed error covariance matrix estimating an error thereof while selecting a plurality of motion model setting methods from a plurality of cases. A multi-target tracking device characterized by the following.