JP2000075023A - Sensor group-controlling device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the tracking performance of a target by allocating each sensor to each target preferentially for the target that is turning when tracking a plurality of targets with a plurality of sensors. SOLUTION: A sensor group-controlling device is provided with bias error evaluation equipment 15 for evaluating the bias constituent of a prediction error according to observation information from a sensor group 5 and tracking information from a tracking filter group 8, random error evaluation equipment 16 for evaluating the random constituent of the prediction error according to the output of the tracking filter group, and observation necessity evaluation equipment 9 for evaluating observation necessity for each of targets 11-1n based on the bias error evaluation equipment and its output, thus increasing the observation necessity of targets that are turning and preferentially allocating sensors 41-4n to them.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a sensor group management apparatus for instructing which sensor should monitor which target when tracking a plurality of targets using a plurality of sensors.

[0002]

2. Description of the Related Art FIG.
FIG. 1 is a block diagram showing a conventional sensor group management device shown in Japanese Patent Publication No. In _{FIG, 1 1, 1 2, ...} , 1 n is the plurality of targets as a tracking target, 2 is the target group that summarizes the plurality of target 1 ₁ to 1 _n. Here, the target 1 ₁ to 1 _n is, for example, an aircraft. 3 ₁ , 3 ₂ ,
.., 3 _m are observation beams for observing the target group 2. 4 ₁ , 4 ₂ ,..., 4 _m are the observation beams 3 _{1 to} 3
a plurality of sensors for observing each target 1 ₁ to 1 _n with _m, 5 is a group of sensors summarizing the plurality of sensors 4 ₁ to 4 _m. Here, a radar device will be considered as a specific example of the sensors 4 ₁ to 4 _m forming the sensor group 5.

[0003] Reference numeral 6 denotes a tracking filter group 8 which combines observation information output from the sensors 4 ₁ to 4 _m of the sensor group 5 and arranges them for each of the targets ₁₁ to 1 _n to be observed. Is an observation information fusion device that outputs to 7 ₁ ,
7 _2, ..., 7 _n receives the observation information outputted from the observation information fusion device 6, a tracking filter for the tracking of each target 1 ₁ to 1 _n. 8 These multiple tracking filter 7 ₁
7 _n is a group of tracking filters.

[0004] Reference numeral 9 denotes an observation necessity evaluator that judges the necessity of observation of each of the targets ₁₁ to 1 _n using the output of the observation information fusion device 6 and the output of the tracking filter group 8 as inputs. 10 as an input the output of the tracking filter group 8, which is a virtual allocator for outputting an evaluation value for the virtual assignment of each sensor 4 ₁ to 4 _m and the target 1 ₁ to 1 _n of the sensors 5. 11
Is the output of the observation information fusion device 6, the output of the tracking filter group 8,
And virtual output as input allocator 10, the positional relationship between the sensors 4 ₁ to 4 _m and each target ₁ 1 to 1 _n, observed each sensor 4 ₁ to 4 _m is the target ₁ 1 to 1 _n This is an observation effect determiner that calculates the effectiveness for.

[0005] 12 determination result of the need to be outputted from the observed need estimator 9, and the efficacy is inputted as the evaluation value for the observed output from the observed effect evaluation determiner 11, the sensors 4 ₁ to 4 each of the observation beam 3 ₁ to 3 _m from _m is an allocator for determining whether to assign to one of the respective target 1 ₁ to 1 _n. Reference numeral 13 denotes a rule for giving an assignment procedure in the assigner 12.

Next, the operation will be described. First, target ₁ 1 to 1 tracking filter 7 ₁ to _7-n is a subject to be observed
_n , and therefore the number is also the target 1 _{1 -1 n}
There are as many as. Tracking filter group 8 receives them when the observation information of the corresponding target 1 ₁ to 1 _n is obtained from the sensor 4 ₁ to 4 _m, performs extrapolation if not obtained. This performs estimation and prediction of the position and velocity of each target 1 ₁ to 1 _n, and outputs the updated tracking information for each target 1 ₁ to 1 _n. The tracking information is output from the tracking filter group 8.

The observation necessity evaluator 9 receives the observation information fused by the observation information fusion unit 6 and the tracking information from the tracking filter group 8, and receives an expected value of a target tracking error, a target position, a distance to the target, Goal direction, goal speed, goal movement,
Perform observation needs assessment of each target 1 ₁ to 1 _n, based on all or part of such importance of the target itself, the target 1 ₁ to 1
Outputs the evaluation value for each _n .

The virtual assigner 10 receives the output of the tracking filter group 8 and, for example, assigns each of the sensors 4 ₁ to 4 _m to each target 11 ₁
The expected value of the tracking error in each combination when virtually assigned to １1 _n is calculated and output.

The observation effect determiner 11 includes an observation information fusion unit 6,
Tracking filter group 8 receives the output of such virtual allocator 10, the target 1 ₁ to 1 distance and sensors 4 ₁ to 4 _m performance of up to _n, due difference between the estimated error variance around the observation, sensors 4 ₁ to 4 to determine an evaluation value for each combination of _m and target 1 ₁ to 1 _n.

[0010] The allocator 12 receives as input the tracking information from the tracking filter group 8 and the evaluation values from the observation necessity evaluator 9 and the observation effect determiner 11, and based on the input tracking information and evaluation values. according to the procedure of allocation given by the rule 13, the observation beam 3 ₁ to 3
_It is determined which _one of the targets 11 to 1 _n is to be assigned to _m , and this is output as assignment information. Note that this assignment information is input to the sensor group 5, and new observation information is obtained.

[0011] Here will be described processing algorithm of the Kalman filter used in general as the tracking filter 7 ₁ to _7-n. Here, the sampling time is t
_k (k = 1, 2,...). The Kalman filter theory, firstly, to express the motion model of the target 1 ₁ to 1 _n by the following equation (1). x _k = Φ (t _k−1 , t _k ) x _k−1 + Γ (t _k−1 , t _k ) w _k−1 (1)

In the above equation (1), x _k is an N-dimensional state vector representing the true values of the motion parameters of the targets 1 ₁ to 1 _{n at} the sampling time t _k , and Φ (t _k−1 , t _k )
Is an N × N transition matrix representing the transition of the state vector x _k from the sampling time t _k−1 to the sampling time t _k . Also, let w _{k be} an s-dimensional driving noise vector at the sampling time t _k, which is a white noise sequence according to the s-variate normal distribution of the mean 0 and the covariance matrix Q _k . The drive noise is actual target 1 ₁ to 1 _n motion, is to excite a disturbance of the state vector when not match the motion model set by the above formula (1). Γ
(T _k−1 , t _k ) is a conversion matrix of the driving noise vector, and N
Xs matrix.

[0013] Here, for example, when tracking the position and velocity of the target ₁ 1 to 1 _n in a rectangular coordinate system xyz, as the motion model of the above formula (1), the following equation (2) to (4)
It can be set as follows. In this case, the driving noise w
_k represents a disturbance input corresponding to acceleration of each axis of x, y, z.
Let it be a dimensional vector. Note that I _{N * N} represents an N × N unit matrix.

[0014]

(Equation 1)

Next, the observation model of the observation device is expressed as the following equation (5). z _k = Hx _k + v _k (5) In this equation (5), z _k is a u-dimensional observation value vector at the sampling time t _k , and H is u ×
N observation matrices. V _k is the sampling time t
_It is a u-dimensional observation noise vector corresponding to the observation value vector z _k at _k , and is assumed to be a white noise sequence according to the u-variate normal distribution of the mean 0 and the covariance matrix R _k . It is assumed that the driving noise vector w _k and the observation noise vector v _k are independent of each other.

Here, for example, when the positions of the targets ₁₁ to 1 _n in the xyz rectangular coordinate system can be obtained as observation values based on the detection results of the observation device, the following observation model of the above equation (5) is used. Equations (6) and (7) can be set.

[0017]

(Equation 2)

The accumulation of the observed value vectors up to the sampling time t _k is represented by the following equation (8). Z _k = {z ₁ , z ₂ , z ₃ ,..., Z _k−1 , z _k } (8)

According to the theory of the Kalman filter, the estimated value ^ x _k of the state vector x _k when the observation value is obtained at the sampling time t _k according to the above model is expressed by the following equations (9) to (13). Is calculated by Where Q _k
Is a parameter of the target ₁ 1 to 1 _n, R _k is the sensor 4 ₁
It is assumed that the parameter is 4 _m . Note that T represents transposition of a matrix.

¨x _k = Φ (t _{k -1} , t _k ) ^ x _{k -1} (9) ^ x _k = ¨x _k + K _k ｛z _k -H¨x _k・ ・ ・ ... ( 10) where _{ P _k = Φ (t _k−1 , t _k ) ^ P _k−1 Φ (t _k−1 , t _k ) ^T + Γ (t _k−1 , t _k ) Q _k−1} ( t _k−1 , t _k ) ^T (11) K _k = ¨P _k H ^T [H¨P _k H ^T + R _k ] ⁻¹ (12) ^ P _k = [IN _{* N} −K _k H] ¨P _k (13)

The estimated value of the state vector x _k when no observation value is obtained at the sampling time t _k ^ x _k
Is calculated by the following equations (14) to (17). This process is generally called extrapolation.

¨x _k = Φ (t _{k -1} , t _k ) ^ x _{k -1} (14) ^ x _k = ¨x _k (15) where ¨P _k = Φ (t _{k-1, t k) ^} P k-1 Φ (t k-1, t k) T + Γ (t k-1, t k) Q k-1 Γ (t k-1, t k) T ··・ (16) ^ P _k = ¨P _k (17)

Here, ^ x _k , ¨x _k , ^ P _k , ¨P _k
Are defined as in the following equations (18) to (21). ^ x _k = E [x _k | Z _k ] (18) ¨x _k = E [x _k | Z _k-1 ] (19) ^ P _k = E [(x _k − ^ x _k ) (x _k − ^ x _k ) ^T | Z _k ] (20) ¨P _k = E [(x _k − ^ x _k ) (x _k − ^ x _k ) ^T | Z _k-1 ] ... (21)

In the equations (18) to (21), E [|] is a symbol representing an averaging operation. Also, ^ x _k
Is x based on observation information Z _k up to sampling time t _k
k is a conditional mean value of corresponds to the smoothing vector estimating the true value of the state vector of the sampling time t _k on the basis of the observation information Z _k to the sampling time t _k. ¨
x _k is the conditional mean value of x _k based on the observation information Z _k-1 to the sampling time t _k-1, the sampling time t _k on the basis of the observation information Z _k-1 to the sampling time t _k-1 Corresponds to the prediction vector obtained by estimating the true value of the state vector at the step S. Further, ^ P _k and ¨P _k are ^ x _k ,
A smoothed error variance matrix and a prediction error covariance matrix representing an error covariance matrix of ¨x _k . Further, x _k in the equation (12) is called a Kalman gain matrix and is an N × u matrix. Note that the smooth vector ^ x _k calculated by the above equations (9) to (17) is an optimal estimated value under the observation information Z _{k in} the sense that the average of the estimation error is zero and the variance is minimized. Become.

[0025] Note that the tracking filter 7 ₁ to _7-n based on the Kalman filter have the following problems. As can be seen from the above Kalman filter equation, the targets 1 ₁ to 1 _n
Prediction error covariance P _k and error covariance ^ P _k, and the calculation of the Kalman gain matrix K _k not reflect the actual observations. Further, when the target 1 ₁ to 1 _n performs miscellaneous motion interwoven and uniform linear motion or turning motion, it is difficult to set the advance motion model. For the above reasons, when the target 1 ₁ to 1 _n makes a movement that does not match the motion model, the Kalman filter is sometimes continues to output the predicted value far from the observations. As a result, towards the sensor 4 ₁ to 4 _m in the direction of the predicted value G1
When trying to observe the ₁ to 1 _n, the sensor 4 ₁ to 4 _m
Target 1 ₁ to 1 _n are not enter the observation range, it occurs such would remove the tracking.

As a method to avoid this problem,
For example, there are those described in JP-A-62-27677 or JP-A-62-27679.
These be those using a Kalman filter by adding the function of detecting the pivoting of the target 1 ₁ to 1 _n by chi-square test, leave the covariance matrix Q _k of the driving noise and large and small two types available, the target the covariance matrix Q _k smaller when the 1 ₁ to 1 _n is determined to have a uniform linear motion, using respectively the covariance matrix Q _k larger when it is determined that the turning motion It has a first method and a second method that has two types of sampling intervals, large and small, and reduces the sampling interval when it is determined that the target is making a turning motion.

However, even with these techniques, the above problem cannot be completely solved. because,
In order to receive the influence of observation noise, Goal 1 ₁ to the drive noise
To set to match the pivoting movement of 1 _n timely manner appropriate value is because it is actually difficult.

The sampling interval is determined by the sensors 4 ₁ to 4
not intended to be any number finely set over a _m capacity of, to the sampling interval is variable, when using the radar as the sensor 4 ₁ to 4 _m, so must be considered as a specification from the design stage of the antenna ,
There is no point in using existing sensors 4 ₁ to 4 _m (radar) having a constant sampling interval.

Furthermore, Madashimo in the case of changing the sampling interval by a single sensor, since the sensor group management apparatus manages a plurality of sensors 4 ₁ to 4 _m in the sensor group 5, different sampling interval, respectively Then, a scheduling problem occurs for each of the sensors 4 ₁ to 4 _m , and determining the sensor assignment as an optimization problem becomes an intractable problem.

[0030]

[Problems that the Invention is to Solve Since conventional sensor group management apparatus is configured as described above, when using the tracking filter 7 ₁ to _7-n based on the Kalman filter, the set motion model and the target 1 ₁ - If actual motion of 1 _n has only to match the expected value of the tracking error the tracking filter 7 ₁ to _7-n is calculated (difference between the true value and the predicted value) is a valid and does not match Kalman filter As mentioned above,
Not only does the difference between the observed value and the predicted value increase, but the observed value is not reflected in the variance value of the tracking error. Although the tracking error actually increases, the expected value of the error is evaluated as a result. because that is going to be, tracking filter 7 ₁ to _7-n
When using a Kalman filter as observation need is evaluated consequently unduly small, the frequency of the sensor 4 ₁ to 4 _m is allocated is insufficient, there is a problem that remove the tracking.

Further, in the Kalman filter with the turning detecting function, if estimates proper driving noise variance in accordance with the extent of deviation from the motion model of the target 1 ₁ to 1 _n, to follow the target 1 ₁ to 1 _n In addition to this, the prediction error variance and the smooth error variance also reflect this, and are output as a relatively large value. However, since they are actually affected by the observation error, the driving noise variance is appropriately estimated. There is also a problem that it is difficult to completely eliminate the danger of missing a tracking due to a mismatch between models.

The present invention has been made in order to solve the above-described problem. The output of each tracking filter is further monitored to detect a target whose tracking is likely to be lost, and to detect the target. It is an object of the present invention to obtain a sensor group management device that can prevent a risk of missing tracking by assigning sensors with higher priority.

[0033]

A sensor group management device according to the present invention comprises an input device for selecting observation information from a sensor group for each target, and a Kalman filter for tracking the target according to the output of the input device. A configured tracking filter group, a bias error evaluator for evaluating a bias component of a prediction error from the input device and an output of the tracking filter group, and a random error evaluator for evaluating a random component of the prediction error from an output of the tracking filter group And an observation necessity evaluator that evaluates the observation necessity of each target based on the outputs of the bias error evaluator and the random error evaluator, and a virtual allocator that virtually determines the assignment from each sensor to each target. An observation effect determiner that receives an output of a tracking filter group and a virtual allocator and determines an evaluation value for each combination of a sensor and a target; Tracking information from the filter group, and on the basis of the observed need evaluator the evaluation value from the observation effect determiner is one and a allocator for determining the allocation of each sensor and each target.

The sensor group management device according to the present invention has a function of evaluating whether or not a target is making a turning motion based on observation information from an input device and tracking information from a tracking filter group. It is what was held in the vessel.

The sensor group management apparatus according to the present invention has a function of calculating and outputting a bias component of the prediction error of each target by smoothing the difference between the predicted position of the target and the observation position using a low-pass filter. It is provided to the bias error evaluator.

The sensor group management apparatus according to the present invention includes a function of evaluating a random component of a target prediction error by using a major axis radius of an error ellipsoid calculated from a target prediction error covariance matrix in a random error evaluator. It is what you have.

According to the sensor group management apparatus of the present invention, the bias component of the prediction error of each target calculated by smoothing the difference between the predicted position of the target and the observation position is input to the sensor specification input from the sensor specification input device. The minimum value of the bias component of the standardized prediction error,
Alternatively, the bias error evaluator has a function of taking a statistic such as an average value or a maximum value and outputting it as a bias error evaluation value for each target.

According to the sensor group management apparatus of the present invention, the area of the target error ellipsoid on the polar coordinate system E-By plane is determined by the observation area of each sensor included in the sensor specifications input from the sensor specification input device. Random error evaluation of each target by taking statistics such as the minimum value, the average value, and the maximum value of the area of the normalized error ellipsoid on the E-By plane The function of outputting as a value is provided to the random error evaluator.

The sensor group management apparatus according to the present invention has a function of determining the observation necessity of each target by taking a weighted average of the outputs from the bias error evaluator and the random error evaluator, to the observation necessity evaluator. It is what you have.

The sensor group management apparatus according to the present invention is provided with an importance evaluator, and is provided with a target position, a distance to the target, a target traveling direction, a target speed, a target movement, and a target attention level. The importance of each target is evaluated based on the part or the whole, and the observation necessity evaluator receives the output from this importance evaluator and the outputs from the bias error evaluator and the random error evaluator and receives them. It has the function of taking the weighted average and determining the observation necessity evaluation value of each target.

The sensor group management apparatus according to the present invention calculates the product of the output of the observation necessity evaluator and the output of the observation effect determiner as the evaluation value for each combination of the target and the sensor, and calculates the evaluation value for the evaluation value. An assigner has a function to determine sensor assignment as a solution to the combination optimization problem.

[0042]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below. Embodiment 1 FIG. FIG. 1 is a block diagram showing a configuration of a sensor group management device according to Embodiment 1 of the present invention. In the figure, 1 ₁ , 1 ₂ ,..., 1 _n are a plurality of targets as observation targets. These goals 1 ₁ to 1 _n is, for example, a flying object other than an aircraft or aircraft. Aircraft also include, depending on their speed, relatively fast jets, relatively slow helicopters, and intermediate propellers. In addition, large aircraft such as jumbo jet aircraft and small aircraft such as Cessna aircraft are included depending on the size. 2 is a target group that summarizes the plurality of target 1 ₁ to 1 _n.

Also, 3 ₁ , 3 ₂ ,..., 3 _m are observation beams for observing the respective targets 1 ₁ to 1 _n of the target group 2.
_{4 1, 4 2, ...,} 4 m is a plurality of sensor for observing the target ₁ 1 to 1 _n by using the observation beam 3 ₁ to 3 _m.
Here, specific examples of the plurality of sensors 4 ₁ to 4 _m, for example, consider a radar apparatus. 5 is a sensor group that summarizes the plurality of sensors 4 ₁ to 4 _m. 1
4 receives the outputs of the sensors 4 ₁ to 4 _m of sensors 5,
They were organized by target 1 ₁ to 1 _n, which is an input device for outputting a corresponding observation information to the tracking filter 7 ₁ to _7-n was of the tracking filter group 8 to be described later.

[0044] 7 _1, 7 _2, ..., 7 _n receives the observation information outputted from the input device 14, each target of the target group 2 ₁ 1 -
There are a plurality of tracking filters that perform tracking of 1 _n . Each of these tracking filters 7 _{1 to} 7 _n has one target 11 ₁ to
Corresponding to 1 _n, and outputs the tracking information of the target ₁ 1 to 1 _n. 8 is a tracking filter group that summarizes the plurality of tracking filter 7 ₁ to _7-n.

Reference numeral 15 denotes a bias error evaluator that receives the output of the input device 14 and the output of each of the tracking filters 7 _{1 to} 7 _n of the tracking filter group 8 and evaluates the bias component of the prediction error. based on the observation information to an output tracking information of the tracking filter group 8, the target 1 ₁ to 1
_n evaluates whether or not the robot performs a turning motion, and a target observation position included in the observation information from the input device 14;
By using a low-pass filter to smooth the difference (residual) from the target prediction position included in the tracking information from the tracking filter group 8, the bias component of the prediction error of each of the targets ₁₁ to 1 _n is calculated and output. Is what you do.

A random error evaluator 16 receives the tracking information output from each of the tracking filters 7 _{1 to} 7 _n of the tracking filter group 8 and evaluates a random component of a tracking prediction error.
Calculating an error ellipsoid from the target prediction position and the prediction error covariance matrix included in the tracking information from the tracking filter group 8;
Based on the length of the major axis radius of the error ellipsoid, each target 1
_It calculates and outputs random components of prediction errors of ₁ to 1 _n .

[0047] 9 receives the output of the bias error evaluator 15 and random error evaluator 16, and the like expected value of turning decision result and the tracking error of the target 1 ₁ to 1 _n, the target 1 ₁ -
This is an observation necessity evaluator that evaluates the necessity of 1 _n observations and outputs the result to an assigner 12 described later.

[0048] 10 receives the output from the tracking filter 7 ₁ to _7-n of the tracking filter group 8, for example, each of the sensors 4 ₁ to 4 _m of sensors 5, the target ₁ 1 to 1 _n of the target group 2 Is a virtual allocator that calculates and outputs an expected value of a tracking error in each combination when virtual allocation is performed. 11 was assigned virtually by the virtual allocator 10, the sensor 4 ₁ for each combination of to 4 _m and the target 1 ₁ to 1 _n, for example, determine the effect of observation and degree of improvement of the tracking accuracy as a measure Then, it is an observation effect determiner that outputs the result to an assigner 12 described later.

[0049] 12 tracking filter group 8, observation need estimator 9, observed in response to an output from the effect determination unit 11, the target 1 ₁ target group 2 of each sensor 4 ₁ to 4 _m of sensors 5 to 1
An allocator that determines allocation to _n .

Next, the operation will be described. First, a theory and a configuration method required for configuring the bias error evaluator 15 and the random error evaluator 16, which are newly provided as inputs to the observation necessity evaluator 9 in the present invention, will be described.

This theory is based on the goal 1 based on the result of the tracking process.
Error between the predicted value and the true value position and velocity of the ₁ to 1 _n, i.e. relates tracking prediction error, in this case, the prediction error of the bias component (hereinafter, referred to as bias error) and random component (hereinafter, random errors ).

First, the random error will be described. In the tracking processing algorithm based on the Kalman filter, at each sampling time t _k , a prediction error covariance matrix ¨P _k is calculated as an error covariance matrix of a prediction vector ¨x _k obtained as a result of tracking. Note that the prediction error covariance matrix kP _k indicates an evaluation value of the random error. The value of the prediction error covariance matrix P _k, the actual motion of the target 1 ₁ to 1 _n is other motion or a motion conforming to the motion model (Constant Velocity) (e.g. pivoting movement, etc.) It does not depend on

The result of the bias error differs depending on whether the targets 1 ₁ to 1 _n are performing a straight-line motion at constant speed or performing another motion such as turning. First, the goals 1 ₁ to 1 _n
It is known that the estimation result by the Kalman filter becomes an unbiased estimator when the robot performs a straight-line motion at a constant speed.
That is, in this case, the bias error becomes zero. On the other hand, when the targets 11 ₁ to 1 _n are performing a turning motion or the like, as shown in FIG. 2, the result of the tracking process (the curve shown by the dotted line) is the actual motion of the targets 11 ₁ to 1 _n (the solid line). (Shown curve), and a predetermined bias error indicated by a thick arrow is generated as a result. The bias error covariance Q _k and # driving noise vector is a parameter defining the characteristics of the tracking filter 7 ₁ to _7-n
It fluctuates depending on the covariance R _{k, r} of the observation noise vector of the r sensor 4 _r and the position of the target.

The predicted vector で obtained as a result of the tracking process
the predicted position of the target ₁ 1 to 1 _n included in x _k, the sensor 4
By smoothing the residual vector, which is the difference from the actual observation value from ₁ to 4 _m , using a low-pass filter, the tracking bias error can be estimated as shown in FIG. FIG. 3A shows the change of the residual vector with respect to the sampling time of the target which shifts from the constant velocity linear motion to the turning motion, and FIG. The change of the bias error with respect to the sampling time is shown. This smoothing is performed using a low-pass filter such as a primary filter or a Kalman filter.

Next, the overall operation of the sensor group management device according to the first embodiment will be described. First, the input device 14 combines observation information output from the plurality of sensors 4 ₁ to 4 _m of the sensor group 5 to obtain a target 11 ₁ to be observed.
整理 1 _n are arranged and output to the tracking filter group 8. That is, the allocator 12 determines which sensor 4 ₁ to 4 _m
Is assigned to which of the goals 1 ₁ to 1 _n
Tracking filter 7 ₁ corresponding to the output of each sensor 4 ₁ to 4 _m
７7 _{n is} output. Also, one goal 1 ₁ to 1 _n
When a plurality of observation beams 3 _{1 to} 3 _m are assigned to each other, the data are fused by averaging the outputs of the respective sensors 4 ₁ to 4 _m or by other calculations, and the data is combined with a tracking filter group. 8 is output. Thus, the input device 14 has a function of linking the plurality of sensors 4 ₁ to 4 _m and a plurality of tracking filters 7 ₁ to _7-n.

Each of the tracking filters 7 ₁ to 7 ₁ of the tracking filter group 8
7 _n is a target 1 corresponding to the input device 14.
Receiving observation information of ₁ to 1 _n, performs estimation and prediction of the position and velocity of the target 1 ₁ to 1 _n, and outputs the updated tracking information for each target 1 ₁ to 1 _n. On the other hand, when the observation information of the target 1 ₁ to 1 _n is not obtained, by performing the extrapolation processing performs estimation and prediction of the position and velocity of each target 1 ₁ to 1 _n, their tracking information Is updated and output. Note that a general Kalman filter is used for these processes.

The bias error evaluator 15 receives outputs from the input device 14 and the tracking filter group 8, and calculates a bias error prediction vector (bias error prediction vector) ¨ξ _k which is a bias component of the prediction error. Magnitude of the calculated bias error prediction vector Kushi _k is output to the observation needs estimator 9 as a bias error evaluation value. Thus, the bias error prediction vector Kushi _k of magnitude to that output as it is, the follow-up delay is large target 1 ₁ to 1
Observation need for _n is greater evaluation, the sensor 4 ₁ to 4 _m is easily assigned.

Further, the random error evaluator 16 receives the output from the tracking filter group 8, calculates an error ellipsoid from the target predicted value ¨x _k and the prediction error variance matrix ¨P _k , and calculates the error ellipsoid. The length of the major axis radius is obtained and output as a random error evaluation value. In equation (22) shown below, the left side is a chi-square distribution with three degrees of freedom, and the region formed by z _{k in} equation (22) using d as a parameter is an ellipsoid. The ellipsoid is called the error ellipsoid. _{[Z k -H¨x k] S k} -1 [z k -H¨x k] T ≦ d ··· (22)

Here, S _k of the above equation (22) is given by the following equation (23), and H in the equation (23) is represented by the following equation (7).
, And R _k is the covariance matrix of the observation noise. S _k = H¨P _k H ^T + R _k (23)

[0060] Incidentally, the error ellipsoid to be smaller when observing the target 1 ₁ to 1 _n, since the increase to be observed, the observed need for long target 1 ₁ to 1 _n of the time not observed It has been acclaimed assigned sensors 4 ₁ to 4 _m is easily performed. Here, the long axis radius of the error ellipsoid is used as the evaluation value, but the volume of the error ellipsoid may be used as the evaluation value.

[0061] Observation need estimator 9 receives the output from these bias error evaluator 15 and random error evaluator 16, the observation needs of each target 1 ₁ to 1 _n by taking the weighted average of their evaluation value And

Here, if it is attempted to determine the necessity of observation by using only one of the output of the bias error evaluator 15 and the output of the random error evaluator 16, the following problem occurs. That is, when you decide the need observed by only the output of the bias error evaluator 15, observation need for target 1 ₁ to 1 _n the vehicle is turning is increased sensor 4 _{1-4
Although m} is more likely to be assigned, there is a possibility that other targets 1 ₁ to 1 _n traveling straight ahead at a constant speed may not be observed at all. On the other hand, when the necessity of observation is determined only by the output of the random error evaluator 16, the turning target 1 ₁
The sensors 4 ₁ to 4 _m are not assigned preferentially to １1 _n .

Therefore, the bias error estimator 15
And by taking weighted averages of the evaluation values output from the random error evaluator 16, respectively, so that the sensors 4 ₁ to 4 _m are preferentially assigned to the turning targets 11 ₁ to 1 _n while the constant speed linear motion is performed. it is possible to assign the sensor 4 ₁ to 4 _m taking an appropriate observation interval to target 1 ₁ to 1 _n that the.

The virtual allocator 10 controls the tracking filter group 8
Position and velocity of each target 1 ₁ to 1 _n which is the output, and inputs these error variance, in the case where for example, assign a respective sensor 4 ₁ to 4 _m in the target 1 ₁ to 1 _n, each combination Calculates and outputs the estimated error variance of the position and velocity at.

The observation effect determiner 11 receives the output of the virtual allocator 10 as an input. If observations are not made, the positions and velocities of the targets 1 ₁ to 1 _n must be used as the estimated values as they are, but if observations are made, the obtained observation information is used as the estimated values. as a highly information, it is possible to obtain the position and velocity of each target 1 ₁ to 1 _n. Therefore, the objectives 1 ₁ to 1 _{n for} observation
Is smaller than the error ellipsoid without observation.

The observation effect determiner 11 calculates the difference or ratio of the volume of the error ellipsoid from the prediction error variance output from the tracking filter group 8 and the prediction error variance output from the virtual allocator 10. determining shrinks degree of error ellipsoid as an effect of observing each target 1 ₁ to 1 _n from the sensors 4 ₁ to 4 _m. In this way, what kind of observation is performed by the virtual assignment by the virtual assigner 10 before the actual assignment is evaluated.

Next, the allocator 12 calculates an allocation evaluation matrix by fusing the output of the observation necessity evaluator 9 and the output of the observation effect determiner 11 by their product, and solves it as a combination optimization problem. There were assigned the sensors 4 ₁ to
Output for 4 _m . For example, a #i sensor (i = 1,
2, ..., m) 4 _i is the #j target (j = 1,2, ..., n) 1
observation effect evaluation value a _ij when assigning to _j and #j target 1
_j of the product C _ij observation needs assessment value b _j calculated by the following equation (24), making the allocation evaluation matrix {C _ij}. C _ij = a _ij × b _j (24)

Here, for example, each of the sensors 4 ₁ to 4 _m observes _one target ₁₁ to 1 _n at the same time, and each of the targets 11 ₁ to 1 _n
When 1 _n is that there is a constraint that is not observed from only one sensor 4 ₁ to 4 _m at the same time, the following equation (25)
By solving as a combination optimization problem for obtaining X _ij that maximizes the evaluation function expression by the following expression (28) based on the constraint expression expressed by (27), the above evaluation value is realized to the maximum. assignment of the sensor 4 ₁ to 4 _m can be obtained.

[0069]

(Equation 3)

X _ij represents an assignment for observing the #j target 1 _j from the #i sensor 4 _i. If this is “1”, this assignment is actually adopted, and if it is “0”, it is not adopted. It means.

[0071] According to the sensor group management apparatus of the first embodiment, and observation need highly evaluated for target 1 ₁ to 1 _n which is not the target 1 ₁ to 1 _n and long observation pivots, each sensor and a degree of contraction of the prediction error caused by performing the observation of the target 1 ₁ to 1 _n from 4 ₁ to 4 _m, evaluate the effect of the combination of the sensors 4 ₁ to 4 _m and each target 1 ₁ to 1 _n -It is possible to determine and determine the sensor assignment.

In the first embodiment, the case where a radar is used as a sensor to track a flying target has been described. However, the present invention is also effective when applied to sensors other than radar. For example, an infrared sensor that detects infrared rays emitted by a target, or an ESM (E
Electronic Warfare Support
Measure) can be considered. However, these sensors can detect a target and measure an azimuth, but cannot measure a distance alone. Therefore, it is necessary to geometrically locate the target position by a combination of a plurality of sensors arranged at different positions. The processing after the target position is obtained in this manner is the same as the case where a radar is used as a sensor. Further, the present invention can be applied to other sensors as long as the target position can be obtained.

As described above, according to the first embodiment, since the bias error evaluator 15 for evaluating the turning of the targets ₁₁ to 1 _n is provided, the turning target has priority over other targets. it becomes possible to assign a sensor 4 ₁ to 4 _m, the residual was smoothed by the bias error evaluator 15, since the calculated bias component of the prediction error, deviation from the motion model with respect to the actual motion The target 11 ₁ to 1 _n having a higher degree of observation has a higher necessity for observation, so that the sensors 4 ₁ to 4 _m can be easily allocated, and the random error evaluator 16 for evaluating the size of the error ellipsoid is used. since comprises, effects such as prevented that ambiguity of the target 1 ₁ to 1 _n will remove the tracking too large without sensor 4 ₁ to 4 _m is assigned long is obtained.

Further, according to the first embodiment, the observation necessity evaluator 9 uses the bias error evaluator 1
5 and a weighted average of the outputs from the random error evaluator 16 and the observation is necessary, so that the sensors 4 ₁ to 4 _m are preferentially assigned to the turning targets ₁₁ to 1 _n . it is possible to assign the sensor 4 ₁ to 4 _m also taking an appropriate observation interval to target 1 ₁ to 1 _n for constant speed straight, the allocator 12, the observed effect judging the output of the observation needs estimator 9 since then reduced to the optimization problem and an output of the vessel 11 are fused by their product, it is possible to determine the allocation of the sensor 4 ₁ to 4 _m so as to maximize the sum of the evaluation values, thus, a plurality of sensors 4 ₁ to 4
It is possible to simultaneously track a plurality of targets 1 ₁ to 1 _n by making the most effective use of _m, and it is possible to use an existing optimization algorithm as a solution to this problem. Can be used to save computation time.

Embodiment 2 FIG. 4 is a block diagram showing a configuration of a sensor group management device according to Embodiment 2 of the present invention. In the figure, ₁₁ to 1 _n are goals, 2 is a goal group, 3
₁ to 3 _m observation beam, 4 ₁ to 4 _m sensors, 5 sensors, 7 ₁ to _7-n are tracking filter, 8 tracking filter group, 9 observations need evaluator, 10 virtual allocator, 1
1 is an observation effect determiner, 12 is an assigner, 14 is an input device, 15 is a bias error evaluator, and 16 is a random error evaluator, which are the same as those shown in FIG. Are the parts corresponding to them.

Reference numeral 17 denotes a plurality of sensors 4 of the sensor group 5.
_{This is} a sensor specification input device for inputting sensor specifications of ₁ to 4 _m to the bias error evaluator 15 or the random error evaluator 16. The second embodiment is different from the sensor group management device of the first embodiment in that the second embodiment includes the sensor specification input device 17 in addition to the above-described units.

The bias error evaluator 15 receives the sensor specifications from the sensor specification input device 17, the observation information from the input device 14, and the tracking information from the tracking filter group 8, and includes them in the observation information. By using a low-pass filter to smooth the residual that is the difference between the target observation position included in the tracking information and the target predicted position included in the tracking information, the bias of the prediction error of each of the targets ₁₁ to 1 _n is reduced. to calculate the components, minimum bias component of the prediction error is normalized by the respective sensors 4 ₁ to 4 _m of each observation area, which is the standardized included it in the sensor specifications received from the sensor specifications input unit 17 This is different from the one denoted by the same reference numeral in FIG. 1 in that a value or a statistic such as an average value or a maximum value is taken and output as a bias error evaluation value for each of the targets ₁₁ to 1 _n . Is

Further, the random error evaluator 16 also uses the tracking information from the tracking filter group 8 and the sensor
7, the area on the polar coordinate system E-By plane of the error ellipsoid obtained from the target prediction position and the prediction error covariance matrix included in the tracking information is input to the sensor specification. Of the observation area of each of the sensors 4 ₁ to 4 _m included in the sensor data received from the detector 17 on the E-By plane, and the normalized error ellipsoid on the E-By plane The same reference numerals are given in FIG. 1 in that the statistical values such as the minimum value, the average value, and the maximum value of the areas are output as the random error evaluation values of the respective targets ₁₁ to 1 _n . Is different from

Next, the operation will be described. In the sensor group management device according to the first embodiment, the bias error evaluator 15 and the random error evaluator 16 estimate the magnitude of the bias error or the random error, respectively.
I was evaluating the size itself. However, in the first embodiment, the magnitude of these bias errors and random errors, sensors 4 ₁ to observe the target 1 ₁ to 1 _n
Depends on the position and accuracy of 4 _m , the sensor 4
Position and accuracy of ₁ to 4 _m is not reflected in the need observed.

Therefore, in the sensor group management device according to the second embodiment, each of the bias error evaluator 15 and the random error evaluator 16 takes into account the sensor specifications input from the sensor 4 ₁ using the relative error size of from to 4 _m are to be evaluated.

The following is an example of the operation of the bias error evaluator 15 and the random error evaluator 16 when a radar is used as the sensors 4 ₁ to 4 _m . First, the procedure for calculating the bias error evaluation value V _k ^b containing the sensor specifications into account in the bias error evaluator 15.

FIG. 5 is an explanatory diagram showing the meaning of each symbol used in the following description. In FIG, r is the radar which is used as #r sensor 4 _r, BW _r is the beam width of the radar _r, BW r / 2 is a half-value width. ¨p _{k, r} is a relative position vector of the predicted position with respect to the radar r, and ¨ξ _k is a predicted vector of a bias component of the tracking prediction error. ¨p ′ _{k, r} is the corrected predicted position obtained by adding the predicted vector ¨ξ _k to the predicted position,
A relative position vector with respect to the radar r, and φ _{k, r} is an angle formed by the two relative position vectors ¨p _{k, r} and ¨p ' _{k, r} .

The following procedure is executed independently for each of the targets ₁₁ to 1 _n . First, the predicted position of the target based on the prediction vector X _k of the target 1 ₁ to 1 _n obtained as output of the tracking filter group 8 is given by H¨x _k. Note that H
Is the observation matrix of equation (7). Here, this predicted position H
¨x the radar r of _{k (r = 1,2, ...,} m) Relative position vector P _k with _{respect, r} is the position vector of the radar r and _{[a r b r c r]} T, the following It is expressed by equation (29). _{¨p k, r = H¨x k -} [a r b r c r] T ··· (29)

On the other hand, a new target predicted position (corrected predicted position) obtained by adding the predicted vector ¨ξ _k of the bias component of the tracking prediction error obtained by smoothing the residual to the predicted position vector H¨x _k. ) Is given by H¨x _k + ¨ξ _k . Therefore, each radar r (r =
1, 2, ..., m) relative position vector {P ' _{k, r}
Is represented by the following equation (30). _{¨p 'k, r = H¨x k} + ¨ξ k - [a r b r c r] T ··· (30)

The angle φ _{k, r} formed by these two relative position vectors ¨p _{k, r} and ¨p ' _{k, r} is _obtained for each radar r by the following equation (31).

[0086]

(Equation 4)

Here, (a, b) represents the inner product of vectors a and b, and {a} represents the Euclidean norm of vector a. Next, a value η _{k, r obtained} by normalizing the absolute value of the angle φ _{k, r} with a half width BW _r / 2 of the beam width BW _r of each radar r is given by:
It is defined by the following equation (32).

[0088]

(Equation 5)

[0089] Finally, the minimum value among all radar r of the normalized value eta _{k, r,} and a bias error evaluation value V _k ^b of the target 1 ₁ to 1 _n, the following equation (33 ).

[0090]

(Equation 6)

Here, the method of selecting the minimum value as a statistic is adopted because the radar 11 having the best condition for tracking maintenance still has the target 11 ₁ to 1 within the observation beam 3 _{1 to} 3 _m . _This is to increase the bias error evaluation value only when it is difficult to keep _n , but an average value or a maximum value may be selected as a statistic. Incidentally, taking the average value average observation beam 3 ₁ caught difficulty each target 1 ₁ into to 3 _m
The bias error evaluation value is 1 _n , and when the maximum value is taken, the evaluation value is such that it is difficult to capture within the observation beam 3 _{1 to} 3 _m when the observation is performed with the radar r having the worst condition for tracking maintenance.

[0092] Next, the following procedure for calculating the random error evaluation value V _k ^r which takes into account the sensor specifications at random error evaluator 16. The idea, the sensor 41 _to the appearance of the error is to observe the target ₁ 1 ~1 _{n
It depends on} the position and accuracy of _m . For example, in the case where even observed at any sensor 4 ₁ to 4 _m is considered that the error is large, it is considered that the observed need for target 1 ₁ to 1 _n evaluates the random error to be large. Therefore, here, to compare the area of the observation beam 3 ₁ to 3 _m in area and radar r error ellipsoid target 1 ₁ to 1 _n in E-By A plane.

First, at the sampling time t _k , a vector θ _{k, r} representing the true value of the target angle viewed from each radar r (r = 1, 2,..., M) is defined as in the following equation (34). I do.

[0094]

(Equation 7)

Using the prediction vector ¨x _k obtained from the tracking filters 7 _{1 to} 7 _n , the angle prediction vector ¨θ
_{k and r} are given by the following equations (35) and (36).

[0096]

(Equation 8)

Here, [ _ar b _{r cr} ] is x of the radar r.
a position vector in the yz coordinate system, [xyz] is the position component of the desired predicted value x _k.

[0098] Further, if the angle prediction vector Shita _k, a mean value of _r and theta ^m _{k, r,} the angle prediction vector Shita _k, random error Shita _k of _{r, r} - [theta] ^m _k, the _r The covariance matrix is given by the following equation (3
It can be expressed as 7). _{A k, r = E [(} ¨θ k, r -θ k, r m) (¨θ k, r -θ k, r m) T] = F k, r ¨P k F k, r T ··・ (37)

Here, F _{k, r} is set as the following equation (38). Further, ¨P _k is a covariance matrix of a random error of the prediction vector, that is, a prediction error covariance matrix.

[0100]

(Equation 9)

Next, the area of the error ellipse is calculated. Variable in the following equation _{(39) [¨θ k, r} -θ k, r m] A k, r -1 [¨θ k, r -θ k, r m] T ··· (39) free According to the chi-square distribution of degree 2, the area of ¨θ _{k, r} that satisfies the following equation (40) with d ′ as a parameter forms an ellipse in the E-By plane. _{[¨θ k, r -θ k,} r m] A k, r -1 [¨θ k, r -θ k, r m] T ≦ d '··· (40)

The area of the error ellipsoid is calculated by the following equation (41).

[0103]

(Equation 10)

[0104] On the other hand, radar r (r = 1,2, ..., m) when the beam width of the BW _r, E-By beamwidth region
The area in the plane is defined by the following equation (42).

[0105]

[Equation 11]

Next, a value ρ _{k, r obtained} by standardizing the area σ _{k, r} of the error ellipsoid with the area s _r ^BW of the beam width region of each radar is given.
Is defined by the following equation (43).

[0107]

(Equation 12)

[0108] Finally, the ρ _{k, r} of all radar r (r =
1,2, ..., the minimum value among m), a random error evaluation value V _k ^r represented by the following formula (44).

[0109]

(Equation 13)

Here, the method of selecting the minimum value as a statistic is adopted because the radar 11 having the best condition for tracking and maintaining the target 11 ₁ to 1 within the observation beam 3 _{1 to} 3 _m is used. _This is to make the random error evaluation value a high value only when it is difficult to keep _n , but an average value or a maximum value may be selected as a statistic. Incidentally, taking the average value average observation beam 3 ₁ caught difficulty each target 1 ₁ into to 3 _m
Becomes a random error evaluation value of 1 _n, also the observation beam 3 ₁ caught difficulty into to 3 _m in the case of performing observation in most conditions poor radar r for Track Maintenance Taking the maximum value as an evaluation value.

As described above, according to the second embodiment, the bias error evaluator 15 and the random error evaluator 16
Respect, each sensor from the sensor specification input unit 17 4 _1-4
Since entering the _m sensors specifications, the sensor 4 ₁ to 4 _m
The necessity of observation in consideration of the position and accuracy of the target is evaluated, whereby the bias error evaluator 15 allows each target 1 ₁
In addition to detecting the turn of 〜1 _n , the bias error generated by the turn can be evaluated including the ease of observation from each of the sensors 4 ₁ to 4 _m . not only calculates the magnitude of the random error of each target 1 ₁ to 1 _n, and can be evaluated, including to the relative magnitude of the random error as seen from the sensors 4 ₁ to 4 _m, observation Since the necessity evaluator 9 determines the necessity of observation in consideration of the sensor specifications, the sensors 4 ₁ to 4 _m are assigned to the targets ₁₁ to 1 _{n which} are likely to lose tracking for the sensor group 5. The effect that it becomes easy is acquired.

Embodiment 3 FIG. 6 is a block diagram showing the configuration of a sensor group management device according to Embodiment 3 of the present invention. The corresponding parts are denoted by the same reference numerals as in FIG. 4 and description thereof is omitted. In FIG, 18 receives the output from the input device 14 and a tracking filter group 8, the target 1 ₁ target group 2
_For each ~ 1 _n , the importance of the target is evaluated based on the distance to the target, the direction of the target, the speed of the target, the movement of the target, the degree of attention of the target itself, and part or all of the arrival time of the target. In the third embodiment, the importance evaluator 18 is added to the sensor group management device according to the second embodiment. Note that the observation necessity evaluator 9 receives the output from the importance evaluator 18 in addition to the output from the bias error evaluator 15 and the output from the random error evaluator 16, and weights these three evaluation values. taking an average, in that evaluating the observed need for each target 1 ₁ to 1 _n, it is different from those designated by the same reference numerals in FIG.

Next, the operation will be described. Here, FIG.
The observation necessity evaluator 9 in the sensor group management device of the first embodiment shown in FIG. 4 or the sensor group management device of the second embodiment shown in FIG. And the random error evaluator 16
By using the random component of the prediction error output from, it had decided observation needs of each target 1 ₁ to 1 _n, the observed need estimator 9 in sensor group management apparatus according to the third embodiment, in addition to the output of the bias error evaluator 15 and random error evaluator 16 is also inputted importance evaluator 18 importance of each target 1 ₁ to 1 _n obtained from.

The importance of each of the targets 1 ₁ to 1 _n is evaluated by the importance evaluator 18 based on the output from the input device 14 and the output from the tracking filter group 8 based on the distance to the target and the target. For example, the following is performed using a part or all of the traveling direction of the target, the speed of the target, the movement of the target, the degree of attention of the target itself, and the arrival time of the target.

1. Distance Goal: sensor 41 _{to when m} or target 1 ₁ to 1 _n which far from certain region is determined not to be critical, the target 1 ₁ to 1 _n is close to the sensor 4 _{1-4 m} etc. Increase the importance of 2. Going direction of the target: If the sensors 4 ₁ to 4 _m or the targets 11 ₁ to 1 _n approaching a specific area are important and the targets 11 ₁ to 1 _n away are not important, the sensor 4 goal ₁ coming toward the _{1 ~4 m} such as 1 ~1 _n
Increase the importance of 3. Target speed: sensors 4 ₁ among to 4 _m or specific area come target 1 ₁ to 1 _n for approaching, if what speed is high is determined to be important, a large target speed 1 ₁ Increase the importance of 重要 1 _n . 4. Goal movement: Goal 11 ₁ that moves fast and is mobile
If 1 _n is determined to be important, its goal 1 ₁
Increase the importance of 重要 1 _n . 5. Attention level of the goal itself: Goal 11 ₁ to 1 _{n by} some means
Size, characteristics, when the kind could be identified, to increase the target 1 ₁ to 1 importance of the determined target 1 ₁ to 1 _n and _n is noteworthy from the identification result. 6. The goal of arrival time: the position of the target 1 ₁ to 1 _n, the traveling direction, by fusing speed, and calculates the time to reach a particular area, with the arrival time is shorter target 1 ₁ to 1 _n is important If it is determined, the importance of the goals ₁₁ to 1 _n is increased.

[0116] Observation need estimator 9, in addition to the output of the bias error evaluator 15 and random error evaluator 16, the importance evaluator 18 each target 1 ₁ to 1 obtained from
The importance of _n is also used as a reference for determining the necessity of observation, and the necessity of observation is determined by taking a weighted average of these three inputs.
Incidentally, the importance index for evaluating the importance of the target 1 ₁ to 1 _n by evaluator 18 is not limited thereto, the target 1 ₁ to 1 _n in some sense as to whether the observed need Any other index can be used as long as the index can be evaluated.

As described above, according to the third embodiment, the importance of each of the targets 1 ₁ to 1 _n obtained by the importance evaluator 18 is also used as an evaluation criterion for the necessity of observation.
Unlike the first and second embodiments in which the sensor group 8 is managed assuming that the values of the targets are all equal, it is possible to objectively evaluate the observation necessity of each of the targets ₁₁ to 1 _n. Goals 11 to ₁ that require special observation
_n , the sensors 4 ₁ to 4 _m can be assigned to the selected targets 11 ₁ to 1 _{n in a} focused manner, and the targets 11 ₁ to 1 _n that require tracking are selected and selected. effect is obtained that it becomes possible to manage the sensors 8 so as not particularly lose sight of the target 1 ₁ to 1 _n has.

[0118]

As described above, according to the present invention, a tracking filter group constituted by a Kalman filter tracks a target in accordance with an output of an input device for selecting observation information from a sensor group for each target. Based on the output of this tracking filter group and the output of the input device in the bias error evaluator,
In addition to evaluating the bias component of the prediction error, the random error evaluator evaluates the random component of the prediction error based on the output of the tracking filter group. Evaluate the observation necessity for each target at, the output of the virtual allocator that virtually determines the allocation from each sensor to each target, and the observation effect determiner that receives the output of the tracking filter group,
An evaluation value is determined for each combination of the sensor and the target, and the assignment of each sensor and each target is determined based on the tracking information from the tracking filter group and the evaluation values from the observation necessity evaluator and the observation effect determiner. Because it was configured to be performed by the allocator,
Even if the set motion model does not match the actual motion of the target, the difference between the observed value and the predicted value increases, and the observed value is not reflected in the variance of the tracking error. Since the expected value of the error will not be underestimated despite the increase, when the Kalman filter is used as the tracking filter, the necessity of observation is consequently underestimated, To detect a target whose tracking is likely to be lost due to insufficient frequency of sensor allocation, and to assign a sensor with priority to that target, thereby preventing the risk of tracking being missed There is an effect that a sensor group management device that can perform the above is obtained.

According to the present invention, the bias error evaluator is configured to evaluate whether or not the target is making a turning motion based on the observation information from the input device and the tracking information from the tracking filter group. There is an effect that it is possible to assign a sensor to a turning target more preferentially than other targets.

According to the present invention, the bias error estimator is configured to evaluate the bias component of the prediction error of each target by smoothing the difference between the target predicted position and the observation position with a low-pass filter. Therefore, there is an effect that the need for observation increases as the degree of deviation of the actual motion from the motion model increases, so that sensors can be more easily assigned.

According to the present invention, the random error estimator is configured to evaluate the random component of the target prediction error using the major axis radius of the error ellipsoid calculated from the target prediction error covariance matrix. In addition, since the sensor is not allocated for a long time, it is possible to prevent the ambiguity of the target from being too large and the tracking from being missed.

According to the present invention, each target is provided with a sensor specification input device to which the sensor specification is input, and the bias error evaluator smoothes the difference between the predicted position of the target and the observation position. The bias component of the prediction error of is normalized by the observation range of each sensor included in the sensor specifications, and the statistic such as the minimum value, the average value, and the maximum value of the bias component of the normalized prediction error is obtained. Since it is configured to output as the bias error evaluation value of each target, it is possible to not only detect the turning of each target, but also evaluate the bias error generated by turning, including the ease of observation from each sensor. This makes it possible to set the observation necessity in consideration of the sensor specifications.

According to the present invention, the sensor specification input device for inputting the sensor specifications is provided, and the area of the target error ellipsoid on the polar coordinate system E-By plane is determined by the random error evaluator. The observation area of each sensor included in the specifications is normalized by the area on the E-By plane, and the standardized error ellipsoid has the minimum, average, and maximum values of the area on the E-By plane. It is configured to take the amount and output it as the random error evaluation value of each target, so not only calculate the magnitude of the random error of each target, but also the relative magnitude of the random error seen from each sensor It is possible to set the observation necessity in consideration of the sensor specifications.

According to the present invention, the observation necessity estimator is configured to take the weighted average of the outputs from the bias error evaluator and the random error evaluator to determine the observation necessity of each target. There is an effect that it is possible to assign a sensor to a target performing a constant-velocity straight-moving at an appropriate observation interval while preferentially assigning a sensor to a target performing a turning motion. .

According to the present invention, the importance of each target is determined based on part or all of the position of the target, the distance to the target, the traveling direction of the target, the speed of the target, the movement of the target, and the degree of attention of the target itself. In addition to providing an importance evaluator that evaluates, the observation necessity evaluator outputs the output from this importance evaluator,
By taking a weighted average of the outputs from the bias error evaluator and the random error evaluator and determining the evaluation value of the observation necessity of each target, it is possible to objectively evaluate the observation necessity of each target. In particular, there is an effect that it is possible to select a target that needs to be observed, and assign a sensor to the selected target in a focused manner.

According to the present invention, the allocator determines an evaluation value for each combination of the target and the sensor from the product of the output of the observation necessity evaluator and the output of the observation effect determiner, and determines the combination for the evaluation value. Since the configuration is such that the sensor assignment is determined as a solution to the optimization problem, the output of the observation necessity evaluator and the output of the observation effect determiner can be fused by the product and reduced to the optimization problem. Since the sensor assignment that maximizes the sum of the values can be determined, not only can multiple targets be used simultaneously to track multiple targets at the same time, but also existing solutions can be used to solve this problem. Since the optimization algorithm can be used, there is an effect that the calculation time can be saved by using the sub-optimization algorithm.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a sensor group management device according to a first embodiment of the present invention.

FIG. 2 is an explanatory diagram showing occurrence of a bias error with respect to a turning target in the first embodiment.

FIG. 3 is an explanatory diagram illustrating calculation of a bias error according to the first embodiment.

FIG. 4 is a block diagram showing a sensor group management device according to a second embodiment of the present invention.

FIG. 5 is an explanatory diagram showing a bias error evaluation method according to a second embodiment.

FIG. 6 is a block diagram showing a sensor group management device according to a third embodiment of the present invention.

FIG. 7 is a block diagram showing a conventional sensor group management device.

[Explanation of symbols]

1 ₁ , 1 ₂ , ..., 1 _n goals, 2 goal groups, 3 ₁ ,
3 ₂ , ..., 3 _m observation beam, 4 ₁ , 4 ₂ , ..., 4 _m
Sensor, 5 sensor group, 6 observation information fusion device, 7 ₁ , 7
₂ ,..., 7 _n tracking filters, 8 tracking filters, 9
Observation necessity evaluator, 10 virtual assigner, 11 observation effect determiner, 12 assigner, 14 input device, 1
5 bias error evaluator, 16 random error evaluator,
17 Sensor specification input device, 18 Importance evaluator.

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Shingo Tsujimichi 2-3-2 Marunouchi, Chiyoda-ku, Tokyo Inside Mitsubishi Electric Corporation (72) Inventor Yoshio Kosuge 2-3-2 Marunouchi, Chiyoda-ku, Tokyo F term in Mitsubishi Electric Corporation (reference) 5J070 AC01 AH40 AH50 AK40 BB16 BB20

Claims (9)

[Claims] 1. A sensor group management device for managing a sensor group composed of a plurality of sensors and tracking a plurality of targets of a target group, comprising: receiving a plurality of pieces of observation information from the sensor group; An input device for selecting, a tracking filter group configured of a plurality of tracking filters based on a Kalman filter that performs the tracking of the target according to the output of the input device, and an output of the input device and an output of the tracking filter group And a bias error evaluator that evaluates the bias component of the prediction error, a random error evaluator that receives the output of the tracking filter group and evaluates a random component of the prediction error, and an output of the bias error evaluator and the An observation necessity evaluator that evaluates the observation necessity of each target based on an output of a random error evaluator; each target of the target group and the sensor A virtual allocator for virtually determining an allocation to each sensor of the group, and an observation effect determiner for determining an observation effect for each allocation between the target and the sensor virtually determined by the virtual allocator. And an allocator that determines allocation of each target of the target group and each sensor of the sensor group based on the output of the observation necessity evaluator and the output of the observation effect determiner. Characteristic sensor group management device. 2. A bias error evaluator for evaluating whether or not a target makes a turning motion based on observation information received from an input device and tracking information received from a tracking filter group. The sensor group management device according to claim 1, wherein: 3. A bias error evaluator smoothes a difference between a target observation position included in observation information received from an input device and a target predicted position included in tracking information received from a tracking filter group by a low-pass filter. 2. The sensor group management device according to claim 1, wherein a bias component of a prediction error of each of the plurality of targets is calculated and output. 4. A random error evaluator calculates an error ellipsoid from a target prediction position and a prediction error covariance matrix included in tracking information received from a tracking filter group, and calculates a length of a major axis radius of the error ellipsoid. 2. The sensor group management device according to claim 1, wherein a random component of a prediction error of each target is evaluated by the evaluation. 5. A sensor specification input device for outputting sensor specifications of each sensor of a sensor group, wherein a bias error evaluator includes a sensor specification from the sensor specification input device and observation information from the input device. And receiving tracking information from the tracking filter group, and smoothing the difference between the target observation position included in the observation information and the target predicted position included in the tracking information with a low-pass filter, thereby predicting each target in the target group. Calculating the bias component of the error, normalizing the bias component of the prediction error in the observation range of each of the sensors included in the sensor specifications, and standardizing the minimum value and the average value of the bias component of the standardized prediction error 2. The sensor group according to claim 1, wherein a statistic such as a maximum value or a maximum value is obtained and output as a bias error evaluation value for each of the targets. Management device. 6. A sensor specification input device for outputting sensor specifications of each sensor of a sensor group, wherein a random error evaluator is configured to perform a sensor specification from the sensor specification input device and a tracking from a tracking filter group. Upon receiving the information, the area of the error ellipsoid obtained from the target predicted position and the prediction error covariance matrix included in the tracking information on the polar coordinate system E-By plane is measured for each of the sensors included in the sensor specifications. The area is normalized by the area on the E-By plane, and a statistic such as a minimum value, an average value, or a maximum value of the area of the normalized error ellipsoid on the E-By plane is obtained, and each of the target values is calculated. The sensor group management device according to claim 1, wherein the sensor group management device outputs the random error evaluation value. 7. An observation necessity evaluator calculates a weighted average of a bias component of a prediction error output from a bias error evaluator and a random component of a prediction error output from a random error evaluator, and calculates a weighted average for each target. 2. The sensor group management device according to claim 1, wherein the sensor group management device evaluates the necessity of observation. 8. For each target in the target group, the position of the target, the distance to the target, the direction of travel of the target, the speed of the target, the movement of the target, and the degree of attention of the target itself are partially or entirely determined. An importance evaluator for evaluating the importance, wherein the observation necessity evaluator receives the output of the importance evaluator, the output of the bias error evaluator, and the output of the random error evaluator, and receives these three evaluation values. 2. The sensor group management device according to claim 1, wherein the observation necessity for each of the targets is evaluated from the weighted average of. 9. An allocator fuses an output of the observation necessity evaluator and an output of the observation effect determiner by a product thereof, and evaluates an evaluation value for each combination of each target of the target group and each sensor of the sensor group. The sensor group management device according to claim 1, wherein a solution obtained by solving a combination optimization problem with respect to the evaluation value is output as sensor assignment.