JP5419797B2 - Target tracking device - Google Patents

Description

  The present invention relates to a target tracking device that generates a target track that is a track of a target by using an observation value of position information obtained from a sensor that observes the target.

  Currently, many papers, patents, and other documents have already been used to track the target using the observation values obtained by observing the target with a sensor. Proposals have been made.

  For example, Patent Literature 1 describes an example of a method for determining resource allocation of a sensor according to a tracking situation. In this prior art, depending on the individual tracking status of the track of the tracking target, such as the size of the error ellipse, the number of detection omissions, the target distance, the target speed, and the distance between the target and the sensor, Set priority every time. Sensor resources are preferentially assigned to a target with a high priority.

  FIG. 9 is a block diagram showing a conventional apparatus as shown in Patent Document 1. In FIG. The configuration in FIG. 9 corresponds to the content described in the column of the prior art in Patent Document 1. Here, Patent Document 1 also describes the control of parameters that determine the operation of the filter and the acquisition of observation conditions such as information on the area where the clutter occurs. However, here, those contents are omitted, and the sensor control function in this conventional apparatus, that is, the function of determining sensor assignment based on the tracking situation will be mainly described.

  The sensor group 100 includes a plurality of sensors such as a first sensor 100a and a second sensor 100b. In addition, the sensor group 100 observes the target designated at the time designated by the operation condition output by the sensor group instruction unit 101, and outputs the observation information to the observation information fusion unit 102. When the observation information fusion unit 102 receives the observation information from any one of the sensors in the sensor group 100, the observation information fusion unit 102 sends the target observation value to the tracking processing unit 103 that estimates the motion specifications of the target from the above operating conditions. Is output.

  The tracking processing unit 103 estimates and predicts the position and speed of each target by executing tracking calculation of the corresponding target, and updates the tracking information of the target. The tracking information extraction unit 104 is information on an error in an estimated value of a target motion specification (for example, position and speed) from target tracking information output by a tracking filter group (not shown) included in the tracking processing unit 103. The error ellipse (error covariance matrix) that is the estimated error range is extracted and the information is output to the target state evaluation unit 105.

  The target state evaluation unit 105 sets an evaluation value corresponding to the priority of sensor allocation for each target. Specifically, since the estimated error range is obtained from the tracking filters belonging to the tracking filter group, the target state evaluation unit 105 sets a large evaluation value for the purpose of reducing the error for a wide target.

  Next, when receiving the evaluation value from the target state evaluation unit 105, the resource allocation method creation unit 106 refers to the tracking database 107, and based on the evaluation value output from the target state evaluation unit 105, the sensor allocation for each target. Multiple sets of assignments and evaluation values for the distribution effect are calculated and output. That is, the resource allocation method creation unit 106 determines a sensor and an observation time to be assigned to the target based on the tracking state of each target that is an evaluation value output from the target state evaluation unit 105. Multiple assignments are made as candidates, and for each of these multiple candidates, the expected value of how much the tracking accuracy of each target is improved by observation of the assigned sensor is weighted by the size of the target state evaluation value The distribution effect evaluation value is calculated.

The degree of improvement in tracking accuracy can be calculated by calculating the error covariance matrix of the Kalman filter. The Kalman filter is characterized in that it calculates not only the estimated value of the target track but also its estimated error. The calculation method of the estimated value and the estimated error is shown below. A smooth value that is an estimated value of the current sampling time k of the target track is denoted by x _{k | k,} and a smooth error covariance matrix that is an estimated error is denoted by P _{k | k} .

When the target track is updated with the observation value z _{k + 1, j} obtained by the sensor at the next sampling time k + 1, first, a prediction process is performed. The prediction processing calculation is performed by calculation using the following equations (1) and (2).

In Equations (1) and (2), Φ _{k + 1} is a transition matrix, and Q _{k + 1} is a drive noise covariance matrix.

Next, smoothing processing is performed using the sensor observation values z _{k + 1, j} and the following equations (3) and (4).

In Equations (3) and (4), K _{k + 1} is a Kalman gain, H _{k + 1} is an observation matrix, and S _{k + 1} is a residual covariance matrix. Further, the Kalman gain K _{k + 1} and the residual covariance matrix S _{k + 1} can be calculated using the following equations (5) and (6).

However, in these equations (5) and (6), R _{k + 1} is an observation error covariance matrix, and is determined from parameters specific to the sensor.

Here, since the smoothing error covariance matrix P _{k + 1 | k + 1} does not depend on the observation values z _{k, j} , the equations (2), (4), (5), and (6) are used before actual observation. Thus, P _{k + 1 | k + 1} is calculated, and the prediction calculation of the effect of the observation can be performed based on the difference from the estimation error P _{k | k} of the current target track.

  Next, the resource management calculation unit 108 receives a plurality of sensor assignments and distribution effect evaluation values for each resource from the resource allocation method creation unit 106. When the resource management calculation unit 108 receives the current operation status of the sensor group 100, the resource management calculation unit 108 refers to the resource database 109 to determine whether or not each sensor can be assigned and the load on the sensor group 100 such as radiant energy. The sensor allocation with a large distribution effect evaluation value is selected in consideration of how much the value becomes.

When the resource management calculation unit 108 determines the optimum distribution method, the sensor group instruction unit 101 instructs the operation of the sensors constituting the sensor group 100 according to the determination.
The above is the main procedure of sensor control by the conventional apparatus.

Japanese Patent No. 4014785

However, the prior art has the following problems.
Here, radar is considered as an example of the sensor. This radar radiates a radio wave toward a target and performs signal processing on the reflected wave to obtain information on the position of the target. For this reason, power is consumed as the number of times of use (number of observations) increases. In the following, a case where radar control (sensor control) is performed so as to minimize the power consumption is considered.

  Here, it is assumed that the target track estimation accuracy at a certain time does not satisfy the required accuracy. In this case, the resource allocation method creation unit 106 and the resource management calculation unit 108 attempt to determine the number of times of use for achieving the required accuracy with the least number of times of use with two radars.

  Assuming that the number of times the first radar is used is n and the number of times the second radar is used is m, in order to predict the effect, n + m times of equations (2), (4), (5), (6) are calculated. Is required. In other words, in order to determine n and m that minimize the number of times of use, n + m times of equations (2), (4), (5), and (6) for all combinations of n and m that are realistic numbers of times of use. It is necessary to determine whether it is possible to achieve the required accuracy from the effect, and to select a combination with the smallest number of times of use from among combinations that can achieve the required accuracy. Therefore, the conventional apparatus has a problem that the amount of calculation increases in accordance with the number of combination candidates in the trial and error.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a target tracking device that can efficiently determine the necessary minimum number of sensor uses.

  The target tracking device of the present invention is for creating a target track by tracking the observation values obtained from a plurality of sensors arranged at different positions, and using the observation values obtained from the sensors. A tracking unit that calculates a target motion specification, derives the target track from the calculated motion specification, and calculates an error covariance matrix of the target track, and an error covariance of the target track A sensor control necessity determination unit that determines whether or not to perform sensor control using an estimation accuracy of the target track based on a matrix and a predetermined required accuracy, an error covariance matrix of the target track, and the sensor A sensor use count determination unit for deriving a sensor use count for each of the plurality of sensors from a constrained optimization problem using the observation accuracy for each and the required accuracy; and the sensor use count determination In accordance with the sensor usage count derived by section, in which and a sensor instruction unit for instructing the observations to each of the plurality of sensors.

  According to the target tracking device of the present invention, the sensor usage count determination unit calculates the sensor usage count for each of the plurality of sensors using the error covariance matrix of the target track and the observation accuracy and required accuracy for each sensor. Since it is derived from a constrained optimization problem, it is possible to reduce the amount of calculation compared to the conventional apparatus, so that the minimum necessary number of sensor uses can be determined efficiently.

It is a block diagram which shows the target tracking apparatus by Embodiment 1 of this invention. It is a flowchart which shows the process sequence of the target tracking apparatus of FIG. It is explanatory drawing for demonstrating the system for determining the frequency | count of radar use. It is a block diagram which shows the target tracking apparatus by Embodiment 2 of this invention. It is a flowchart which shows the process sequence of the target tracking apparatus of FIG. It is explanatory drawing for demonstrating the system for determining the frequency | count of radar use. It is a block diagram which shows the target tracking apparatus by Embodiment 3 of this invention. It is a flowchart which shows the process sequence of the target tracking apparatus of FIG. It is a block diagram which shows the conventional apparatus as shown to patent document 1. FIG.

Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings.
Embodiment 1 FIG.
FIG. 1 is a block diagram showing a target tracking device (multi-target tracking device) according to Embodiment 1 of the present invention.
In FIG. 1, a target tracking device 10 according to the first embodiment is connected to a radar group 1 as a sensor group including a first radar (first sensor) 1a and a second radar (second sensor) 1b. The first and second radars 1a and 1b are installed at different locations.

  The target tracking device 10 includes a tracking processing unit 2, a radar control necessity determination unit 3 as a sensor control necessity determination unit, and a radar use number determination unit as a sensor use number determination unit (sensor observation number determination unit). 4 and a radar instruction unit 5 as a sensor instruction unit.

  The target tracking device 10 usually observes at a predetermined time interval using the first and second radars 1a and 1b and is obtained periodically from the first and second radars 1a and 1b. The target tracking process is performed by using the observation value to generate the target track.

  Further, the target tracking device 10 derives an estimated value of the target track by performing a calculation using a tracking filter (Kalman filter or the like) on the observed value during the tracking process. Since the estimated value of the target wake reflects information on the observed values of the past several times of the radar group 1, the estimated error is an observation error that is an error included in the observed value of one observation of the radar group 1. Smaller than.

  Furthermore, it is assumed that the target tracking device 10 stores in advance a predetermined required accuracy (target track accuracy requested by the user) for the target track set by the user. Further, when the Kalman filter is used as the tracking filter, the target tracking device 10 stores the observation accuracy of the observation value of the radar group 1 and can calculate the estimation accuracy which is the accuracy of the estimated value of the target track. Furthermore, the target tracking device 10 performs radar control (sensor control) when it is found that the estimation accuracy does not satisfy the required accuracy and the difference between the two is large, depending on the target tracking status. Focus radar resources on specific targets and areas in a short period of time.

  Next, each method of the target tracking by the functional blocks 2 to 5 of the target tracking device 10 and the determination regarding the radar control associated therewith will be described. FIG. 2 is a flowchart showing a processing procedure of the target tracking device 10 of FIG. This flowchart shows the flow of processing for one observation time of observation values periodically, and each functional block 2 to 5 performs the entire processing of FIG. 2 once for each observation. Hereinafter, the processing contents of tracking and radar control determination will be described with reference to this flowchart.

  In FIG. 2, first, in the “normal tracking process” in step S1, an observation value obtained by periodic observation of at least one of the first and second radars 1a and 1b is input to the tracking processing unit 2, The tracking processing unit 2 performs target tracking processing using the observed value. In the target tracking process, the motion error of the target is calculated by reducing the observation error included in the observed value. The procedure will be described below.

When the smooth value of the target wake before processing by the observed value is x _{k−1 | k−1} and the error covariance matrix is P _{k−1 | k−1} , the tracking processing unit 2 uses the following formula: Prediction processing is performed by performing calculations using (7) and (8).

However, in these formulas (7) and (8), Φ _k is a transition matrix, and Q _k is a drive noise covariance matrix.

  Next, the tracking processing unit 2 determines whether or not the observation value can be associated with the target track. Specifically, the tracking processing unit 2 determines that association is possible when the following equation (9) is satisfied when the observation value obtained by the radar group 1 is used.

However, in this equation (9), d is a gate size parameter used in chi-square test. H _k is an observation matrix. Further, S _k is a residual covariance matrix and can be calculated according to the following equation (10).

However, in this equation (10), R _k is an observation error covariance matrix, and can be calculated from the observation error standard deviation of the distance of the first or second radar 1a, 1b and the observation error standard deviation of the angle.

The tracking processing unit 2 performs the smoothing process using the following equations (11) and (12) when the equation (9) is established and it is determined that the target track and the observed value can be associated with each other. X _{k | k} which is an estimated value of the latest motion parameters of the target and P _{k | k} which is an error covariance matrix thereof are obtained.

However, in these formulas (11) and (12), K _k is a Kalman gain and can be calculated as the following formula (13).

Next, in “determination of necessity of radar control” in step S2, it is determined whether the radar control necessity determination unit 3 should temporarily increase the frequency of use of a specific radar. Here, the standard deviations of the estimation errors between the x-axis direction and the y-axis direction of the target track are assumed to be p _x0 and p _y0 , respectively. The radar control necessity determination unit 3 converts the error covariance matrix P _{k | k} of the target wake into xy coordinates, and converts it into a diagonal term component corresponding to xx of the converted error covariance matrix and yy. P _x0 and p _y0 are approximated by calculating the square root with the corresponding diagonal term component, respectively.

Further, the standard deviation of the x-axis direction component and a y-axis direction component of the specified estimation error by required accuracy p _{X_d,} and p _{Y_d.} The radar control necessity determination unit 3 performs the radar control necessity determination using the standard deviation of these errors. In step S3, the radar control necessity determination unit 3 determines that the difference between the currently estimated error standard deviation of the target track (error standard deviation in the estimated accuracy) and the error standard deviation in the requested accuracy is the threshold _Δth . If it exceeds, that is, if any of the following equations (14) and (15) is satisfied, the usage frequency of at least one of the first and second radars 1a and 1b should be temporarily increased ( Radar control is necessary), and the radar usage count determination unit 4 is caused to execute the process of step S4.

  On the other hand, if any of the conditions of the previous equations (14) and (15) is not satisfied in step S3, the radar control necessity determination unit 3 determines that the radar control is unnecessary, and a series of operations. Exit. Therefore, the radar control necessity determination unit 3 determines whether to perform sensor control using the target track estimation accuracy and the required accuracy.

Next, in “determination of the number of radar use” in step S4, the radar use number determination unit 4 determines the number of use of each radar 1a, 1b (the number of observations imposed on each radar 1a, 1b). The standard deviations of observation errors in the x-axis direction and the y-axis direction of observation values obtained by one observation of the first radar 1a are assumed to be σ _x1 and σ _y1 .

These standard deviations σ _x1 , σ _y1 are the observation error covariance matrix of the coordinates composed of the distance direction axis and the angle direction axis, which are obtained from the distance observation error and the angle observation error of the first radar 1a. It is obtained by the square root of the diagonal term of the covariance matrix obtained by converting to coordinates. Similarly, σ _x2 and σ _y2 are standard deviations of observation errors in the x-axis direction and the y-axis direction of observation values obtained by one observation of the second radar 1b. These standard deviations σ _x1 , σ _y1 , σ _x2 , and σ _y2 correspond to the observation accuracy of each radar 1a and 1b.

The standard deviation of the estimation error of the target track after the accuracy has been improved by the observation of the first radar 1a n times and the second radar 1b of the continuous m times, that is, the standard deviation corresponding to the required accuracy is P _{x_nm} , P _{y_nm.} And These can be approximated as the following formulas (16) and (17) when the observation interval of continuous n times and continuous m times by the first and second radars 1a and 1b is close to zero.

  In addition, since the estimation error of the target track after the accuracy has been improved needs to be smaller than the estimation error based on the required accuracy, it is possible to set a constraint condition such as the following equations (18) and (19).

  Furthermore, since n and m must be positive, a constraint condition such as the following equation (20) can be set.

  Here, it is assumed that the usage times n and m of the first radar 1a and the second radar 1b are determined for the purpose of minimizing the sum. That is, the objective function shown in the following equation (21) is determined to be minimized.

Here, the constraint conditions of the previous equations (18) to (20) and the constraint condition and objective function of the constrained optimization problem consisting of the objective function of equation (21) are expressed in two-dimensional coordinates of (n, m). When expressed in space, it is as shown in FIG. The shaded portion in FIG. 3 is a region that satisfies the constraint conditions, and the lines indicated by the numerical values in parentheses in FIG. 3 are the boundaries of the constraint conditions of equations (18) to (20), respectively. To minimize the objective function, the intersection point in the lower left circle of this region may be selected. The coordinates (n _f , m _f ) of this point can be calculated by simultaneous equations relating to n and m consisting of the boundaries of equations (18) and (19), and the solution is as shown in the following equation (22): become.

The solution to the optimization problem of Equation (22) is not always an integer. For this reason, a minimum integer exceeding n _f and m _f may be used as the number of times the radar is actually applied. Further, the number of hits per observation time of each of the first and second radars 1a and 1b may be adjusted to realize the number of times of use corresponding to the values of n _f and m _f including the decimal part.

  In this example, the configuration assuming two radars, the first and second radars 1a and 1b, has been described. However, the configuration can be easily expanded to three or more radars. In this case, the number of simultaneous equations increases in accordance with the number of radars, but the optimum number of times of use of each radar can be derived by matrix calculation.

Further, when using three or more radars, there is a case where it is desired to emphasize reduction of the number of times of use of a specific radar. In this case, adjustment can be easily made by multiplying the objective function by a coefficient. For example, assuming that N radars are used and the number of use of radar i is n _i , an objective function such as the following equation (23) can be used.

However, in this equation (23), α _i is a weighting coefficient relating to the number of times the radar i is used. When it is desired to particularly reduce the number of times the radar j is used, α _j may be set to a value larger than the weighting coefficient for the number of times other radars are used.

When deriving n _f and m _f , it is assumed that the n _f times of use of the radar 1a and the m _f times of the radar 1b can be completed in a sufficiently short time from the current time. Since the target position changes after a lapse of time, the standard deviations σ _x1 , σ _y1 , σ _x2 , and σ _y2 of the observation errors of the radars change accordingly. Therefore, when either n _f or m _f is a large number, the error due to approximation increases, and the difference between the estimation of the effect of using the radar and the effect of using the actual radar increases. In such a case, error estimation by trial and error according to the prior art may be performed. That is, for n _f and m _f derived by the procedure described in the first embodiment,
max (n _f , m _f )> _{nu_lim}
If the above holds, it is assumed that trial and error according to the prior art is performed. Here, _{nu_lim} is a parameter set in advance.

Next, the "radar instruction" in step S5, the radar instructing unit 5 to perform the target observations n _f times for first radar 1a, also, m _f times the target relative to the second radar 1b Send instructions (commands) through the communication network to make observations.

Next, in the “re-tracking process” in step S6, the tracking processing unit 2 performs the target track tracking process using the observation values obtained by the radars 1a and 1b through the radar instruction in step S5. The processing per observation value is the same as the equations (7) to (13) of “normal tracking processing” in step S1, and the tracking processing unit 2 repeats this processing n _f + m _f times.

  According to the target tracking apparatus of the first embodiment as described above, the radar usage number determination unit 4 uses the error covariance matrix of the target track, the observation accuracy and the required accuracy for each radar, and The number of times the radar is used for each of the two radars 1a and 1b is derived from a constrained optimization problem. This makes it possible to reduce the amount of calculation compared to the conventional device that determines the number of radar use while calculating the error ellipse by trial and error, so that the minimum necessary number of radar use can be determined efficiently. .

Embodiment 2. FIG.
In the first embodiment, the example in which the constraint conditions in the constrained optimization problem are set for the target track estimation accuracy and the required accuracy has been described. On the other hand, in the second embodiment, an example will be described in which the constraint condition in the constrained optimization problem is set for the number of times of radar use (number of times of sensor use).

  FIG. 4 is a block diagram showing a target tracking apparatus according to Embodiment 2 of the present invention. The basic processing content of the target tracking device 20 of the second embodiment is the same as that of the target tracking device 10 of the first embodiment. The target tracking device 20 includes functional blocks 22 to 25 similar to the functional blocks 2 to 5 of the target tracking device 10 according to the first embodiment. Other configurations are the same as those in the first embodiment.

  Next, each method of the target tracking by each functional block 22-25 of the target tracking apparatus 20 and the determination regarding the radar control accompanying it will be described. FIG. 5 is a flowchart showing a processing procedure of the target tracking device 20 of FIG. This flowchart shows the flow of processing for one observation time period of observation values. Each functional block 22 to 25 performs the entire processing of FIG. 5 once for each observation. Hereinafter, the processing contents of tracking and radar control determination will be described with reference to this flowchart.

  In FIG. 5, first, in the “normal tracking process” in step S21, an observation value obtained by periodic observation of at least one of the first and second radars 1a and 1b is input to the tracking processing unit 22, The tracking processing unit 22 performs target tracking processing using the observed value. In the target tracking process, the motion error of the target is calculated by reducing the observation error included in the observed value. The procedure will be described below.

When the smooth value of the target track before processing based on the observed value is x _{k−1 | k−1} and the error covariance matrix is P _{k−1 | k−1} , the tracking processing unit 22 uses the following formula ( Prediction processing is performed by performing calculations using 24) and (25).

In Equations (24) and (25), Φ _k is a transition matrix, and Q _k is a drive noise covariance matrix.

Next, the tracking processing unit 22 determines whether or not the observation value can be associated with the target track. Specifically, when the observation value obtained by the radar group 1 is z _{k, j} , the tracking processing unit 22 determines that the association is possible when the following expression (26) is satisfied.

However, in this formula (26), d is a gate size parameter used in chi-square test. H _k is an observation matrix. S _k is a residual covariance matrix and can be calculated according to the following equation (27).

However, in this equation (27), R _k is an observation error covariance matrix, and can be calculated from the observation error standard deviation of the distance of the first or second radar 1a, 1b and the observation error standard deviation of the angle.

The tracking processing unit 22 performs smoothing processing using the following equations (28) and (29) when it is determined that the equation (26) is established and the target track and the observed value can be associated with each other, X _{k | k} that is an estimated value of the latest motion specification of the target and P _{k | k} that is an error covariance thereof are obtained.

However, in these formulas (28) and (29), K _k is a Kalman gain and can be calculated as the following formula (30).

Next, in “determination of necessity / unnecessity of radar control” in step S22, it is determined whether or not the radar control necessity determination unit 23 should temporarily increase the usage frequency of a specific radar. Here, the standard deviations of the estimation errors between the x-axis direction and the y-axis direction of the target track are assumed to be p _x0 and p _y0 , respectively. The radar control necessity determination unit 23 converts the error covariance matrix P _{k | k} of the target track into xy coordinates, and converts it into a diagonal term component corresponding to xx of the converted error covariance matrix and yy. P _x0 and p _y0 are approximated by calculating the square root with the corresponding diagonal term component, respectively.

Further, the standard deviation of the x-axis direction component and a y-axis direction component of the specified estimation error by required accuracy p _{X_d,} and p _{Y_d.} The radar control necessity determination unit 23 determines the necessity of radar control using the standard deviation of these errors. In step S23, the radar control necessity determination unit 23, the difference between the error standard deviation in the required accuracy (error standard deviation in estimation accuracy) error standard deviation between the target track currently estimated exceeds a threshold delta _th In this case, that is, when any of the following equations (31) and (32) is satisfied, the frequency of use of at least one of the first and second radars 1a and 1b should be temporarily increased (radar) Control is required), and the radar usage count determination unit 24 is caused to execute the process of step S24.

  On the other hand, if any of the conditions of the previous equations (31) and (32) is not satisfied in step S23, the radar control necessity determination unit 23 determines that the radar control is not necessary, and a series of operations. Exit. Accordingly, the radar control necessity determination unit 23 determines whether to perform sensor control using the target track estimation accuracy and the required accuracy.

Next, in “determining the number of times of radar use” in step S24, the radar usage number determination unit 24 determines the number of times each radar 1a, 1b is used (the number of observations that can be imposed on each radar 1a, 1b). The standard deviations of observation errors in the x-axis direction and the y-axis direction of observation values obtained by one observation of the first radar 1a are assumed to be σ _x1 and σ _y1 .

These standard deviations σ _x1 and σ _y1 are the xy observation error covariance matrix, which is obtained from the distance observation error and the angle observation error of the first radar 1a, and is composed of the axis of the distance direction and the axis of the angle direction. It is obtained by the square root of the diagonal term of the covariance matrix obtained by converting to coordinates. Similarly, σ _x2 and σ _y2 are standard deviations of observation errors in the x-axis direction and the y-axis direction of observation values obtained by one observation of the second radar 1b. These standard deviations σ _x1 , σ _y1 , σ _x2 , and σ _y2 correspond to the observation accuracy of each radar 1a and 1b.

_Let P _{x_nm} and P _{y_nm be} the standard deviations of the estimation error of the target track after the accuracy has been improved by observing the first radar 1a n times continuously and the second radar 1b continuously m times. These can be approximated as the following formulas (33) and (34) when the observation interval of continuous n times and continuous m times by the first and second radars 1a and 1b is close to zero.

Here, it is assumed that there is an upper limit (N _max ) in the sum of the number of uses n and m of the first radar 1a and the second radar 1b. In this case, a constraint condition such as the following equation (35) can be set.

  Since n and m must be positive, a constraint condition such as the following equation (36) can be set.

  Here, the number of uses n and m of the first radar 1a and the second radar 1b is determined for the purpose of minimizing the area of the error ellipse after use. This is equivalent to minimizing an objective function such as the following equation (37).

Here, the constraint condition and the objective function of the constrained optimization problem consisting of the constraint condition of the previous expressions (35) and (36) and the objective function of the expression (37) are expressed in two-dimensional coordinates of (n, m). When expressed in space, FIG. 6 is obtained. The shaded portion in FIG. 6 is a region that satisfies the constraint conditions, and the lines indicated by the parenthesized numerical values in FIG. 6 are the boundaries of the constraint conditions in the equations (35) and (36), respectively. To minimize the objective function, the intersection point in the lower left circle of this region may be selected. The coordinates (n _f , m _f ) of this point can be calculated by simultaneous equations relating to n and m consisting of the boundaries of equations (35) and (36).

The solution to this optimization problem is not always an integer. For this reason, a minimum integer exceeding n _f and m _f may be used as the number of times the radar is actually applied. Further, the number of hits per observation time of each of the first and second radars 1a and 1b may be adjusted to realize the number of times of use corresponding to the values of n _f and m _f including the decimal part.

When deriving n _f and m _f , it is assumed that the n _f times of use of the radar 1a and the m _f times of the radar 1b can be completed in a sufficiently short time from the current time. Since the target position changes after a lapse of time, the standard deviations σ _x1 , σ _y1 , σ _x2 , and σ _y2 of the observation errors of the radars change accordingly. Therefore, when either n _f or m _f is a large number, the error due to approximation increases, and the difference between the estimation of the effect of using the radar and the effect of using the actual radar increases. In such a case, error estimation by trial and error according to the prior art may be performed. That is, for n _f and m _f derived by the procedure described in the second embodiment,
max (n _f , m _f )> _{nu_lim}
If the above holds, it is assumed that trial and error according to the prior art is performed. Here, _{nu_lim} is a parameter set in advance.

Next, in the “radar instruction” in step S25, the radar instruction unit 25 performs n _f target observations with respect to the first radar 1a and m _f targets with respect to the second radar 1b. Send instructions (commands) through the communication network to make observations.

Next, in the “re-tracking process” in step S26, the tracking processing unit 22 performs the target track tracking process using the observation values obtained by the radars 1a and 1b through the radar instruction in step S25. The processing per observation value is the same as the formulas (24) to (30) of the “normal tracking processing” in step S21, and the tracking processing unit 22 repeats this processing n _f + m _f times.

  According to the target tracking device of the second embodiment as described above, the radar usage count determination unit 24 sets the constraint condition in the constrained optimization problem for the radar usage count, and the first and second radars 1a and 1b. The number of times each radar is used is derived from the constrained optimization problem (resulting in a constrained optimization problem to obtain an optimal solution). As a result, even when the number of radar uses is limited, the number of radar uses for increasing the effect of reducing the estimation error of the target track can be determined with a small amount of calculation.

  In the first and second embodiments, examples in which the radars 1a and 1b and the radar group 1 are used as sensors and sensor groups have been described. However, the sensor and the sensor group are not limited to the radar and the radar group, respectively, and may be any sensors that can observe the target. In this case, each of the functional blocks 2 to 5 and 22 to 25 in the first and second embodiments may be processed for a sensor other than the radar.

Embodiment 3 FIG.
In the first and second embodiments, the example in which the number of times of use (the number of observations) of each of the first and second radars 1a and 1b is derived from the constrained optimization problem has been described. On the other hand, in the third embodiment, the number of hits is derived from the constrained optimization problem on the premise that the required accuracy is achieved by one observation of the first and second radars 1a and 1b. An example will be described.

  FIG. 7 is a block diagram showing a target tracking apparatus according to Embodiment 3 of the present invention. The basic processing content of the target tracking device 30 of the third embodiment is the same as that of the target tracking device 10 of the first embodiment. The target tracking device 30 includes functional blocks 32, 33, and 35 that are the same as the functional blocks 2, 3, and 5 of the target tracking device 10 of the first embodiment. Further, the target tracking device 30 has a radar hit number determination unit 34 instead of the radar usage number determination unit 4 of the first embodiment. Other configurations are the same as those in the first embodiment.

  Next, each method of the target tracking by the functional blocks 32 to 35 of the target tracking device 30 and the determination regarding the radar control associated therewith will be described. FIG. 8 is a flowchart showing a processing procedure of the target tracking device 30 of FIG. This flowchart shows the flow of processing for one observation time periodically of the observed values, and each functional block 32 to 35 performs the entire processing of FIG. 8 once for each observation. Hereinafter, the processing contents of tracking and radar control determination will be described with reference to this flowchart.

  In FIG. 8, first, in the “normal tracking process” in step S31, an observation value obtained by periodic observation of at least one of the first and second radars 1a and 1b is input to the tracking processing unit 32, The tracking processing unit 32 performs target tracking processing using the observed value. In this target tracking process, noise is removed from the observed values and the target motion specifications are calculated. The procedure will be described below.

When the smooth value of the target track before processing based on the observed value is x _{k−1 | k−1} and the error covariance matrix is P _{k−1 | k−1} , the tracking processing unit 32 uses the following equation: Prediction processing is performed by performing calculations using (38) and (39).

In Equations (38) and (39), Φ _k is a transition matrix, and Q _k is a drive noise covariance matrix.

Next, the tracking processing unit 32 determines whether or not the observation value can be associated with the target track. Specifically, when the observation value obtained by the radar group 31 is z _{k, j} , the tracking processing unit 32 determines that the association is possible if the following expression (40) is satisfied.

However, in this formula (40), d is a gate size parameter used in chi-square test. H _k is an observation matrix. Further, S _k is a residual covariance matrix and can be calculated according to the following equation (41).

However, in this equation (41), R _k is an observation error covariance matrix, and can be calculated from the observation error standard deviation of the distance of the first radar 1a or the second radar 1b and the observation error standard deviation of the angle.

The tracking processing unit 32 performs the smoothing process using the following formulas (42) and (43) when the formula (40) is established and it is determined that the target track and the observed value can be associated with each other. X _{k | k} that is an estimated value of the latest motion specification of the target and P _{k | k} that is an error covariance thereof are obtained.

However, in these formulas (42) and (43), K _k is a Kalman gain and can be calculated as the following formula (44).

Next, in the “radar control necessity determination” in step S32, it is determined whether or not the radar control necessity determination unit 33 should temporarily increase the number of hits of a specific radar. Here, the standard deviations of the estimation errors between the x-axis direction and the y-axis direction of the target track are assumed to be p _x0 and p _y0 , respectively. The radar control necessity determination unit 33 converts the error covariance matrix P _{k | k} of the target track into xy coordinates, and converts the converted error covariance matrix to diag and the diagonal term component corresponding to xx. P _x0 and p _y0 are approximated by calculating the square root with the corresponding diagonal term component, respectively.

Further, the standard deviation of the x-axis direction component and a y-axis direction component of the specified estimation error by required accuracy p _{X_d,} and p _{Y_d.} The radar control necessity determination unit 33 performs the radar control necessity determination using the standard deviation of these errors. In step S33, the radar control necessity determining unit 33, the difference between the error standard deviation in the required accuracy (error standard deviation in estimation accuracy) error standard deviation between the target track currently estimated exceeds a threshold delta _th In this case, that is, when any of the following equations (45) and (46) is satisfied, the number of hits of at least one of the first and second radars 1a and 1b should be temporarily increased (radar) Control is required), and the radar hit number determination unit 34 is caused to execute the process of step S34.

  On the other hand, if any of the conditions of the previous equations (45) and (46) is not satisfied in step S33, the radar control necessity determination unit 33 determines that the radar control is not necessary and performs a series of operations. Exit. Therefore, the radar control necessity determination unit 33 determines whether to perform sensor control using the target track estimation accuracy and the required accuracy.

Next, in “determining the number of radar hits” in step S34, the radar hit number determination unit 34 determines the number of hits of each radar 1a, 1b (the number of hits that can be imposed on each radar 1a, 1b). The standard deviations of observation errors in the x-axis direction and the y-axis direction of observation values obtained by one observation of the first radar 1a are assumed to be σ _x1 and σ _y1 . These are covariances obtained by converting an observation error covariance matrix of coordinates composed of a distance direction axis and an angle direction axis obtained by the distance observation error and the angle observation error of the first radar 1a into xy coordinates. It is obtained by the square root of the diagonal term of the variance matrix. Therefore, these standard deviations σ _x1 and σ _y1 depend on the number of hits H ₁ of the first radar 1a.

Similarly, assuming that the standard deviations of the observation errors in the x-axis direction and the y-axis direction of the observation values obtained by one observation of the second radar 1b are σ _x2 and σ _y2 , these standard deviations σ _x2, sigma _y2 depends on the number of hits of _{H 2} second radar 1b.

Once Observation by hits H ₁ of the first radar 1a, the standard deviation of the estimation error of the target track after the accuracy is improved by a single observation by the number of hits of H ₂ second radar _1b P _{x_H,} P _{Let y_H} . These standard deviations P _{x — H} and P _{y — H} can be approximated as the following equations (47) and (48) when the observation interval is close to zero.

  Here, since it is necessary to make the estimation error of the target track after the accuracy improved smaller than the estimation error based on the required accuracy, it is possible to set a constraint condition such as the following equations (49) and (50). .

Further, since H ₁ and H ₂ must be positive, a constraint condition such as the following equation (51) can be set.

Here, the hit numbers H ₁ and H ₂ of the first radar 1a and the second radar 1b are determined for the purpose of minimizing the power consumption of the two radars. That is, the objective function of the following formula (52) is determined to be minimized. In the following equation (52), P (H) is the power consumption necessary for the radar to realize the hit number H.

The constraint condition and the objective function of the constrained optimization problem consisting of the constraint conditions of the previous expressions (49) to (51) and the objective function of the expression (52) are expressed from the boundary of the constraint conditions (49) and (51). It can be calculated by simultaneous equations relating to H ₁ and H ₂ .

Next, the "radar instruction" in step S35, the radar instruction unit 35 to perform the target observed by hits H ₁ relative to the first radar 1a, also hits H ₂ relative to the second radar 1b Send an instruction (command) through the communication network to perform target observation.

  Next, in the “re-tracking process” in step S36, the tracking processing unit 32 performs the target track tracking process using the observation values obtained by the radars 1a and 1b through the radar instruction in step S35. The processing per observation value is the same as the formulas (38) to (44) of “normal tracking processing” in step S31, and the tracking processing unit 32 repeats this processing twice.

  According to the target tracking device of the third embodiment as described above, the radar hit number determination unit 34 uses the error covariance matrix of the target track and the required accuracy, and the first and second radars 1a and 1b. The number of radar hits for each is derived from the constrained optimization problem (a constrained optimization problem is set and its optimal value is obtained). As a result, the amount of calculation can be reduced as compared with the conventional apparatus, so that the minimum necessary number of radar hits can be determined efficiently.

  DESCRIPTION OF SYMBOLS 1 Radar group (sensor group), 1a 1st radar (1st sensor), 1b 2nd radar (2nd sensor), 2, 22, 32 Tracking processing part, 3, 23 Radar control necessity judgment part (sensor control required , 4, 24 Radar usage count determination unit (sensor usage count determination unit), 5, 25 Radar instruction unit (sensor instruction unit) 10, 20, 30 Target tracking device, 33 Radar control necessity determination unit, 34 Radar hit number determination unit, 35 Radar instruction unit.

Claims (5)

A target tracking device for creating a target track by tracking observation values obtained from a plurality of sensors arranged at different positions,
A tracking process for calculating a target motion specification using an observation value obtained from the sensor, deriving the target track from the calculated motion specification, and calculating an error covariance matrix of the target track And
A sensor control necessity determination unit that determines whether to perform sensor control using the target track estimation accuracy based on the target track error covariance matrix and a predetermined required accuracy;
Using the error covariance matrix of the target wake, the observation accuracy for each sensor, and the required accuracy, the number of times the sensor is used for deriving the number of times the sensor is used for each of the plurality of sensors from the constrained optimization problem A decision unit;
A target tracking device comprising: a sensor instruction unit that instructs each of the plurality of sensors to observe according to the sensor use number derived by the sensor use number determination unit. The sensor usage count determining unit
Set the constraint conditions in the constrained optimization problem for each of the required accuracy and the current estimated accuracy of the target track,
The target tracking device according to claim 1, wherein an objective function in a constrained optimization problem is set as a sum of the number of times the sensor is used. The sensor usage count determining unit
Set the constraint conditions in the constrained optimization problem for each of the required accuracy and the current estimated accuracy of the target track,
2. The objective according to claim 1, wherein the objective function in the constrained optimization problem is set as a sum of values obtained by multiplying the number of times the sensor is used for each of the plurality of sensors by a weighting coefficient for each sensor. Tracking device. The sensor usage count determining unit
Set the constraint conditions in the constrained optimization problem for the number of times the sensor is used,
The target tracking device according to claim 1, wherein an objective function in a constrained optimization problem is set with an estimation accuracy of the target track after use of the sensor. A target tracking device for generating a target track by tracking observation values obtained from a plurality of radars arranged at different positions,
Tracking processing for calculating a target motion specification using observation values obtained from the radar, deriving the target track from the calculated motion specification, and calculating an error covariance matrix of the target track And
A radar control necessity determination unit that determines whether to perform radar control, using the target track estimation accuracy based on the target track error covariance matrix and a predetermined required accuracy;
Using the error covariance matrix of the target wake and the required accuracy, the number of hits at the time of observation of each of the plurality of radars, a radar hit number determination unit that derives from a constrained optimization problem,
A target tracking device comprising: a radar hit number instruction unit that instructs each of the plurality of radars to observe according to the hit number derived by the radar hit number determination unit.

Images