JP2011169751A - Tracking device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a tracking device for obtaining an exact tracking result. <P>SOLUTION: The tracking device includes an observation data management section 12 for selecting observation data at which observation time is used for estimation based on a track establishing determination result, a selection time state candidate generation section 13 for setting all selected observation times as reference times and generating the track candidate at each reference time as a state vector candidate, a state space rough retrieval section 14 for calculating the likelihood ratio of the state vector candidates and retrieving the state vector candidate of the largest likelihood ratio at each reference time, a state space detailed retrieval section 15 for retrieving the state vector estimated value of the largest likelihood ratio at each reference time by iterative calculation using, as the initial value, the state vector candidate of the largest likelihood ratio at each reference time, a state correlation processing section 16 for determining the state vector estimated value of the greatest likelihood at the newest time from the correspondence of the state vector estimated value of the largest likelihood ratio at each reference time, and a track establishing determination section 22 for determining the establishment of a track based on the state vector estimated value. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a tracking device that tracks a moving target using observation data from a sensor as an input, and more particularly to a tracking device that tracks a track of a moving target by maximum likelihood estimation.

  A device using ML-PDA (Maximum Likelihood-Probabilistic Data Association) is known as a tracking device for estimating the most probable track of a moving target using observation data from a sensor (see, for example, Non-Patent Document 1). ).

The tracking device using the ML-PDA estimates the track of the moving target by the maximum likelihood estimation method. First, a track of a moving target is expressed as a state vector x _k such as a target position and speed at a certain time (hereinafter referred to as a reference time). Next, the state vector is extrapolated forward or backward according to a previously assumed target motion, and the probability of the state vector based on the observation data group Z is expressed as a likelihood ratio function φ (Z, x _k ). Then, the target wake is estimated by obtaining a global maximum point x _k ^ML of the likelihood ratio function φ (Z, x _k ) from the obtained observation data.

  Hereinafter, a conventional tracking device using the ML-PDA will be described with reference to FIGS. 13 and 14. FIG. 13 is a block diagram illustrating a configuration of a conventional tracking device using an ML-PDA. FIG. 14 is a flowchart showing the operation of a conventional tracking device using ML-PDA.

  In FIG. 13, a conventional tracking device 1X using ML-PDA includes an observation data storage unit 91, an observation data management unit 92, a state candidate generation unit 93, a state space rough search unit 94, and a state space detailed search. A section 95 and a wake establishment determination section 96 are provided.

  First, in step 901, the observation data storage unit 91 acquires observation data from the sensor, and in the next step 902, the observation data storage unit 91 stores the acquired observation data for each observation time.

Next, in step 903, the observation data management unit 92 selects an observation data group at each time from among the observation data groups stored in the observation data storage unit 91 for the state space rough search unit 94 and the state space detailed search. Whether to input to the unit 95 is selected based on the wake establishment determination result of the wake establishment determination unit 96. Hereinafter referred to as the observation time at which you select from the time t ₁ to t _N.

  Next, in step 904, the state candidate generation unit 93 appropriately sets the reference time, generates a large number of state vector candidates at the reference time, and outputs this state vector candidate group to the state space approximate search unit 94. .

Next, in step 905, the state space approximate search unit 94 receives the state vector candidate group input from the state candidate generation unit 93 and the observation data from time t ₁ to t _N selected by the observation data management unit 92. Using the group, one x _k ^* that maximizes the likelihood ratio function φ (Z, x _k ) is searched from the state vector candidates, and is output to the state space detailed search unit 95. Here, it is assumed that x _k ^* is sufficiently close to the global maximum point x _k ^ML of φ (Z, x _k ).

Next, in step 906, the state space detailed search unit 95 uses the state vector candidate x _k ^* input from the state space approximate search unit 94 as an initial point, and the times t ₁ to t _N selected by the observation data management unit 92. Using the observation data group up to and applying the iterative processing method that converges to x _k ^ML under the above assumption, and the convergence points x _k ^** and φ (Z, x _k ^** ) are sent to the wake establishment determination unit 96 Output. As this iterative processing method, many methods such as the steepest descent method and the quasi-Newton method have been proposed.

In step 907, the wake establishment determination unit 96 determines wake establishment according to predetermined determination conditions, and outputs the determination result to the observation data management unit 92. In addition, the wake establishment determination unit 96 outputs the input state vector estimated value x _k ^** regardless of the determination result of the wake establishment.

By the way, the state candidate generation unit 93 generates state vector candidates based on a position range that can be observed by a sensor, an assumed speed direction, and the like. However, the number of candidates is generally enormous. However, using the property that the likelihood ratio function φ (Z, x _k ) becomes higher as the position of the state vector is closer to the observation data, the result of mapping the observation data to the state space is used as the position component of the state vector candidate. (For example, refer nonpatent literature 2).

  FIG. 15 is a block diagram showing a configuration of another conventional tracking device using an ML-PDA that uses observation data as position components of state vector candidates.

  In FIG. 15, the tracking device 1Y is a conventional tracking device 1X using the ML-PDA in FIG. 13 except that the observation data group selected by the observation data management unit 92 in the state candidate generation unit 93A is input. Is the same.

  The tracking device 1Y uses the observation data as the position component of the state vector candidate in the state candidate generation unit 93A. However, if the set reference time is different from the observation time, the state vector at the observation time is determined in combination with the velocity component candidates, and the state vector is extrapolated forward or backward to the reference time according to the pre-supposed target motion. A thing is generated as a state vector candidate.

  In this way, in the tracking device 1Y, by narrowing down the range to be searched by the state space approximate search unit 94, the number of state vector candidates to be generated is changed to the state vector candidates generated by the state candidate generation unit 93 of the tracking device 1X. The number can be less than the number, and the search can be speeded up.

T. Kirubarajan and Y. Bar-Shalom, Low observable target motion analysis using amplitude information, IEEE Transactions on Aerospace and Electronic Systems, vol.32, no.4, pp.1367-1384, 1996. WR Blanding, PK Willett, Y. Bar-Shalom, and RS Lynch, Directed subspace search ML-PDA with application to active sonar tracking, IEEE Transactions on Aerospace and Electronic Systems, vol.44, no.1, pp.201-216 , 2008.

  As described above, in the conventional tracking device using the ML-PDA, it is necessary to set the reference time in advance. This reference time can be arbitrarily determined, for example, the latest time is set.

  By the way, since the observation environment for obtaining observation data from the sensor, that is, whether the target is observed or the target observation error varies from time to time, the state vector estimation error varies depending on the reference time. Therefore, it is necessary to obtain a reference time that minimizes the estimation error.

  However, in the conventional tracking device using the ML-PDA, since the reference time is fixedly set, the estimation error is not always minimum at the reference time, and the basis for setting the reference time is ambiguous. was there.

  Furthermore, since the observation environment of the sensor is unknown, there is a problem that it is difficult to obtain a reference time that minimizes the estimation error.

In the conventional tracking device using the ML-PDA, the state vector candidate x _k ^* searched by the state space approximate search unit 94 is the global maximum point x _k ^ML of the likelihood ratio function φ (Z, x _k ). Is close enough. The iterative processing method in the state space detailed search unit 95 is effective only under the above assumption. For this reason, when a state vector candidate x _k ^* contrary to the above assumption is obtained, the iteration converges to the local maximum point of the likelihood ratio function by the iterative processing method, and the estimation error increases. there were.

  The present invention has been made to solve the above-described problems, and by eliminating the ambiguity of the basis for setting the reference time and eliminating the state vector estimation value converged to the local maximum point, An object is to obtain a tracking device capable of obtaining an accurate tracking result.

  The tracking device according to the present invention includes an observation data accumulation unit that accumulates observation data input from a sensor for each observation time, and a wake establishment determination result among observation data that is accumulated in the observation data accumulation unit for each observation time. The observation data management unit that selects which observation data is used for estimation based on the observation data, and all the observation times selected by the observation data management unit are set as reference times, and wake candidates at each reference time are selected. The likelihood ratio of the state vector candidate is calculated using the selected time state candidate generation unit to be generated as a state vector candidate and the observation data selected by the observation data management unit, and the likelihood ratio at each reference time is A state space rough search unit for searching for the largest state vector candidate, and a state vector with the maximum likelihood ratio at each reference time searched by the state space rough search unit. A state space detailed search unit that searches for a state vector estimated value with the maximum likelihood ratio at each reference time by iterative calculation using the observation data selected by the observation data management unit as an initial value, A state correlation processing unit for determining the most probable state vector estimated value from the correspondence relationship of the state vector estimated value having the maximum likelihood ratio at each reference time, and a wake from the most probable state vector estimated value at the latest time And a wake establishment determination unit that outputs the wake establishment determination result.

  According to the tracking device according to the present invention, it is possible to obtain an accurate tracking result by eliminating the ambiguity of the basis for setting the reference time and eliminating the state vector estimation value converged to the local maximum point. There is an effect.

It is a block diagram which shows the structure of the tracking apparatus which concerns on Embodiment 1 of this invention. It is a flowchart which shows operation | movement of the tracking apparatus which concerns on Embodiment 1 of this invention. It is a figure which shows the state vector estimated value for every reference | standard time of the tracking apparatus which concerns on Embodiment 1 of this invention. It is a figure showing the result of the prediction process of the tracking apparatus which concerns on Embodiment 1 of this invention, and a gate inside / outside determination process. It is a figure which shows the grouping process result of the state vector estimated value of the tracking apparatus which concerns on Embodiment 1 of this invention. It is a figure which shows the smoothing process result and track determination result of the state vector estimated value for every group of the tracking apparatus which concerns on Embodiment 1 of this invention. It is a block diagram which shows the structure of the tracking apparatus which concerns on Embodiment 2 of this invention. It is a flowchart which shows operation | movement of the tracking apparatus which concerns on Embodiment 2 of this invention. It is a block diagram which shows the structure of the tracking apparatus which concerns on Embodiment 3 of this invention. It is a flowchart which shows operation | movement of the tracking apparatus which concerns on Embodiment 3 of this invention. It is a block diagram which shows the structure of the tracking apparatus which concerns on Embodiment 4 of this invention. It is a flowchart which shows operation | movement of the tracking apparatus which concerns on Embodiment 4 of this invention. It is a block diagram which shows the structure of the conventional tracking apparatus using ML-PDA. It is a flowchart which shows operation | movement of the conventional tracking apparatus using ML-PDA. It is a block diagram which shows the structure of another conventional tracking apparatus using ML-PDA which uses observation data as a positional component of a state vector candidate.

  Hereinafter, preferred embodiments of the tracking device of the present invention will be described with reference to the drawings.

Embodiment 1 FIG.
A tracking device according to Embodiment 1 of the present invention will be described with reference to FIGS. FIG. 1 is a block diagram showing the configuration of the tracking device according to Embodiment 1 of the present invention. In the following, in each figure, the same reference numerals indicate the same or corresponding parts.

  1, the tracking device 1 according to Embodiment 1 of the present invention includes an observation data storage unit 11, an observation data management unit 12, a selected time state candidate generation unit 13, a state space rough search unit 14, and a state. A space detailed search unit 15, a state correlation processing unit 16, and a wake establishment determination unit 22 are provided.

  In FIG. 1, the state correlation processing unit 16 includes a prediction processing unit 17, a gate inside / outside determination unit 18, a grouping processing unit 19, a smoothing processing unit 20, and a wake determination unit 21.

  Next, the operation of the tracking device according to the first embodiment will be described with reference to the drawings.

  FIG. 2 is a flowchart showing the operation of the tracking device according to Embodiment 1 of the present invention.

  First, in step 101, the observation data storage unit 11 acquires observation data from a sensor.

  Next, in step 102, the observation data storage unit 11 stores the acquired observation data for each observation time.

  Next, in step 103, the observation data management unit 12 selects an observation data group at each time among the observation data groups stored in the observation data storage unit 11 at the selected time state candidate generation unit 13, the state space outline. Whether to output to the search unit 14, the state space detailed search unit 15, and the state correlation processing unit 16 is selected based on the track establishment determination result of the track establishment determination unit 22.

Next, in step 104, selects the time state candidate generating unit 13, all from the time sequence t ₁ selected at observation data management unit 12 to t _N is set as the reference time, the state vector candidates per the reference time Is generated and output to the state space approximate search unit 14. However, in order to reduce the calculation load, the observation data mapped to the state space is used as the position component of the state vector candidate.

Next, in step 105, the state space approximate search unit 14 inputs a state vector candidate group from the reference times t ₁ to t _N input from the selected time state candidate generation unit 13, and the reference times t ₁ to t _N. The likelihood ratio function φ (Z, x _k ) is maximized among the state vector candidates using the observation data group from the time t ₁ to t _N selected by the observation data management unit 12 at each time until. Search from x ₁ ^* to x _N ^* one by one, and output to state space detailed search unit 15.

Next, in step 106, the state space detailed search unit 15 sets the state vector candidates x ₁ ^* to x _N ^* from the reference times t ₁ to t _N input from the state space approximate search unit 14 as initial values, respectively. For each time from the reference time t ₁ to t _N , the likelihood ratio function φ (Z, x is obtained by the iterative processing method using the observation data group from the time t ₁ to t _N selected by the observation data management unit 12. _k ) from state vector x ₁ ^** to x _N ^** which maximizes the state, and outputs state vector estimated values x ₁ ^** to x _N ^** to the state correlation processing unit 16.

  The functions of the selection time state candidate generation unit 13, the state space rough search unit 14, and the state space detailed search unit 15 described above correspond to a case where a plurality of reference times are set for each function in the conventional tracking device. This can be realized by repeating the operation of each function in the conventional tracking device by the number of set reference times.

  Next, details of the function of the state correlation processing unit 16 will be described with reference to FIGS.

  FIG. 3 is a diagram showing state vector estimation values for each reference time of the tracking device according to Embodiment 1 of the present invention. Moreover, FIG. 4 is a figure showing the result of the prediction process of the tracking apparatus which concerns on Embodiment 1 of this invention, and a gate inside / outside determination process. FIG. 5 is a diagram showing a grouping process result of the state vector estimation value of the tracking device according to Embodiment 1 of the present invention. Furthermore, FIG. 6 is a figure which shows the smoothing process result and track determination result of the state vector estimated value for every group of the tracking apparatus which concerns on Embodiment 1 of this invention.

First, in step 107, the prediction processing unit 17 performs state vector estimation values x ₁ ^** to x _N ^* from the reference times t ₁ to t _N input from the state space detailed search unit 15 as shown in FIG. ^* until sequentially, as shown in FIG. 4, forward extrapolation to the latest time t _N in accordance with the target motion previously assumed. Here, an extrapolated state vector estimated value x _i ^{** up} to time t _k (where 1 ≦ i <k ≦ N) is referred to as state vector predicted value x _{i, k} ^** .

Next, in step 108, the gate inside / outside determination unit 18, as shown in FIG. 4, the state vector estimated value x _k ^** at a certain reference time t _k and the state vector predicted value x _i obtained by the prediction processing unit 17. _{, K} ^** (where 1 ≦ i <k ≦ N) is evaluated by distances d _{i, k} based on a predetermined matrix C _k as shown in equation (1). If the distance d _{i, k} is less than or equal to a predetermined threshold value d _G , the state vector estimated value x _k ^** and the state vector predicted value x _i ^** have a correspondence, otherwise, the distance d _{i, k} Is larger (longer) than the predetermined threshold value d _G, it is determined that the state vector estimated value x _k ^** and the state vector predicted value x _i ^** have no correspondence. This determination is performed for all of the reference times t ₁ to t _N.

Next, in step 109, the grouping processing unit 19 determines that the state vector estimated values x ₁ ^** to x _N ^** are determined to be compatible by the gate inside / outside determining unit 18 as shown in FIG. Collectively as one group Γ _j . Here, it is assumed that the state vector estimated values x ₁ ^** to x _N ^** are divided into groups Γ ₁ to Γ _n .

Next, in step 110, as shown in FIG. 6, the smoothing processing unit 20 processes the state vector estimation values belonging to the groups Γ ₁ to Γ _n with a filter, so that the state at the latest time t _N of each group. X _n ^*** is obtained from the vector smooth value x ₁ ^*** .

However, in general, the time of the state vector to which a certain group Γ _j belongs is obtained by removing some of the times t ₁ to t _N. The estimated value of the state vector at the removed time is obtained by extrapolating forward the state vector estimated value at the immediately preceding time in accordance with the target motion assumed in advance.

In step 111, the wake determination unit 21 selects the observation vector management unit 12 from the state vector smooth values x ₁ ^*** to x _n ^*** at the latest time t _N from the groups Γ ₁ to Γ _n. The likelihood ratio function φ (Z, x _k ) is calculated using the observed data group from time t ₁ to t _N, and the state vector smooth value x _j ^{*** of the} group having the largest likelihood ratio function value. And a likelihood ratio function φ (Z, x _j ^*** ) to determine a wake, a state vector smooth value x _j ^*** , a likelihood ratio function φ (Z, x _j ^*** ), and Is output to the wake establishment determination unit 22.

However, when there are a plurality of groups having the largest likelihood ratio function value, the state vector smooth values x _j ^*** and φ (Z, x _j ^{*) of the} group having the largest number of state vector smooth values belonging to the group ^{. **} ) is selected and output to the wake establishment determination unit 22.

Finally, in step 112, the wake establishment determination unit 22 determines wake establishment according to a predetermined determination condition, and outputs the determination result to the observation data management unit 12. Further, the wake establishment determination unit 22 outputs the state vector smooth value x _j ^*** input from the state correlation processing unit 16 as an estimated value regardless of the determination result of the wake establishment.

  As described above, according to the first embodiment, the state vector is estimated by setting all the times of the observation data used for estimating the state vector as the reference time, thereby reducing the ambiguity of the basis for setting the reference time. It can be solved.

  In addition, by estimating state vectors at a plurality of different reference times and determining their correspondences by grouping processing, state vector estimates that converge to a local maximum point at a certain reference time are extracted, and such states It is possible to eliminate the vector estimation value.

  Furthermore, by processing the state vector estimated values at a plurality of different reference times with a filter, an effect of further suppressing an increase in estimation error is obtained.

In the first embodiment, it has been described that the distance d _{i, k} between the state vector estimated value x _k ^** and the state vector predicted value x _{i, k} ^** is given by equation (1) in step 108. , Position components Hx _k ^** , Hx _{i, k} ^** (H is an observation matrix, which represents a mapping from the state space to the observation data space), and an expression based on a predetermined matrix D _k It can also be given by the distance d _{i, k} ′ in (2). In this case, the size of the matrix to be calculated becomes smaller than that of the equation (1), and the calculation time can be shortened.

In the first embodiment, in step 110, it is described that a filter for the calculation of the state vector smoothing value at the latest time t _N of each group, the state vector estimate of the most recent time included in each group the basis can also be determined by extrapolating to the latest time t _N in accordance with the target motion previously assumed. In this case, the calculation time can be shortened by omitting complicated filter calculation.

  In the first embodiment, the state space rough search unit 14 searches for one state vector candidate that maximizes the likelihood ratio function at each reference time, and the state space detailed search unit 15 sets them as initial values. As described above, it is assumed that the convergence point is searched one by one by applying the iterative processing method, but finally K convergence points are obtained and the correspondence relationship of the obtained K × N state vector estimation values is determined. However, the convergence point is not limited to one. However, by limiting the number of convergence points to one, the size of the correspondence determination problem is reduced, and the calculation time can be shortened.

Embodiment 2. FIG.
A tracking device according to Embodiment 2 of the present invention will be described with reference to FIGS. FIG. 7 is a block diagram showing the configuration of the tracking device according to Embodiment 2 of the present invention.

  In FIG. 7, the configuration of tracking device 1A according to the second embodiment of the present invention further includes a reference time setting unit 23 and a state vector estimated value storage unit 24, as compared with the configuration of the first embodiment. Is different.

  Next, the operation of the tracking device according to the second embodiment will be described with reference to the drawings.

  FIG. 8 is a flowchart showing the operation of the tracking device according to Embodiment 2 of the present invention. Steps 201-203, 205-207, 209-212, 215, and 216 in FIG. 8 correspond to steps 101-112 in FIG.

  In FIG. 8, the operation of the tracking device 1A according to the second embodiment of the present invention is compared with the operation of the first embodiment described above, instead of the “loop at the reference time” in step 104-105. Reference point setting) and step 208 (state vector estimated value storage) are added, and a conditional branch (steps 213 and 214) is added between step 212 and step 215. . Therefore, the operation of the tracking device 1A according to the second embodiment will be described below with a focus on these differences.

First, in step 204, the reference time setting unit 23, for the time sequence t ₁ selected at observation data management unit 12 to t _N, first set the time t ₁ to the reference time. When the wake determination unit 21 does not determine a wake, t ₂ , t ₃ ,. . . , And advance the reference time one by one. Hereinafter, the reference time set by the reference time setting unit 23 is referred to as “current reference time”, and a reference time that is earlier than the current reference time is referred to as “past reference time”.

  Next, in step 205 to step 207, the selection time state candidate generation unit 13, the state space approximate search unit 14, and the state space detailed search unit 15 perform a state space search on the current reference time, and perform a gate inside / outside determination unit 18 is output.

  In step 208, the state vector estimated value storage unit 24 stores the state vector estimated value at the current reference time.

  Next, in step 209, the prediction processing unit 17 forwardly extracts the state vector estimated value group at the past reference time accumulated in the state vector estimated value storage unit 24 to the current reference time according to the target motion assumed in advance. Insert a state vector prediction value group.

  Next, Step 210, Step 211, and Step 212 are the same as the operations in the first embodiment except that the operations are for the state vector estimated values at the past and current reference times.

  Next, in step 213, the wake determination unit 21 determines that the wake determination is possible if there is a group in which the number of state vector smooth values belonging to the group is equal to or more than M in advance. In step 215, a state vector smooth value of a group having M or more state vector smooth values is selected, and a wake is determined.

On the other hand, in step 214, the number of cases that are not determined to be track determining the current reference time, if equal to the latest time t _N selected in observation data management unit 12, in step 215, the state vector smoothing values belonging to the group The state vector smooth value of the group with the largest number is selected and the wake is determined.

Further, track determining unit 21, when determining the track, forward extrapolated status vectors smoothed value of the selected group to the latest time t _N.

  As described above, in the second embodiment, the reference time is sequentially set from the observation data time sequence, the state vectors are sequentially estimated with respect to the reference time, and the correspondence of the state vector estimated values becomes M or more. If it does, it is not estimated at the subsequent reference time. As a result, in the first embodiment described above, it is possible to reduce the calculation load rather than setting the time series of the observation data to the reference time and estimating the state vector simultaneously with respect to the reference time.

In the second embodiment, in step 210, as in the first embodiment, the distance between the state vector estimated value x _k ^** and the state vector predicted value x _{i, k} ^** is given by equation (1). Although described as a thing, it cannot be overemphasized that it can also give by Formula (2).

Embodiment 3 FIG.
A tracking device according to Embodiment 3 of the present invention will be described with reference to FIGS. FIG. 9 is a block diagram showing the configuration of the tracking device according to Embodiment 3 of the present invention.

  In FIG. 9, the configuration of the tracking device 1B according to the third embodiment of the present invention is the latest time between the state correlation processing unit 16 and the wake establishment determination unit 22 as compared with the configuration of the first embodiment. The difference is that a state vector estimated value storage unit 25 and a latest time state correlation processing unit 26 are further provided.

  Next, the operation of the tracking device according to the third embodiment will be described with reference to the drawings.

  FIG. 10 is a flowchart showing the operation of the tracking device according to Embodiment 3 of the present invention. Steps 301-311 and 318 in FIG. 10 correspond to steps 101-112 in FIG.

  In FIG. 10, the operation of the tracking device 1B according to the third embodiment of the present invention is compared with the operation of the first embodiment described above, between step 311 and step 318, step 312 (latest time state vector estimated value). The difference is that steps from (memory) to step 317 (track determination) are added. However, except for step 312 (storage of latest time state vector estimated value), steps 313 to 317 are the same as steps 107 (prediction processing) to step 111 (track determination) of the first embodiment. . Therefore, the operation of the tracking device 1B according to the third embodiment will be described below with a focus on these differences.

  First, by the operations from step 301 to step 311, the tracking device 1B outputs the state vector estimated value at the latest time, as in the configuration and operation of the first embodiment.

  Next, in step 312, the latest time state vector estimated value storage unit 25 stores the latest time state vector estimated value. Further, based on the wake establishment determination result of the wake establishment determination unit 22, it is determined how many state vector estimated values for the past time are stored, and the past state vector estimated values for the time are updated to the latest time state vector estimation. Delete from the value storage unit 25.

  Next, in steps 313 to 317, the operation of the latest time state correlation processing unit 26 is the same as the operation of the state correlation processing unit 16 from step 107 to step 111 as described above.

However, the input value in the state correlation processing unit 16, a state vector estimate from the time t ₁ from the reference time t ₁ obtained from observed data up to t _N to t _N, these state vector estimate, use The time of all observed data is the same. On the other hand, the input value in the latest time state correlation processing unit 26 is the state vector estimated value of the past and latest time output from the state correlation processing unit 16, and these state vector estimated values are the time strings of the observation data used. Is different. That is, the latest time state correlation processing unit 26 determines a correspondence relationship between state vector estimated values from observation data of different time sequences.

  As described above, according to the third embodiment, state vectors are estimated from observation data of a plurality of different time sequences, and the correspondence between them is determined by grouping processing. Even in the case where the convergence to the local maximum point cannot be eliminated by the estimated estimation, it is possible to eliminate such a state vector estimated value.

In the third embodiment, in step 308 and step 314, as in the first embodiment, the distance between the state vector estimated value x _k ^** and the state vector predicted value x _{i, k} ^{** is} expressed by equation (1). However, it is needless to say that it can also be given by equation (2).

  In the third embodiment, it has been described that the configuration and operation of the first embodiment are used in steps 301 to 311. However, the configuration of the second embodiment and the operation of FIG. 8 are used. However, the same effect can be obtained.

Embodiment 4 FIG.
A tracking device according to Embodiment 4 of the present invention will be described with reference to FIGS. FIG. 11 is a block diagram showing the configuration of the tracking device according to Embodiment 4 of the present invention.

  In FIG. 11, the configuration of the tracking device 1 </ b> C according to the fourth embodiment of the present invention is the same as the configuration of the first embodiment described above, except that the configuration in which the wake determination unit 21 is removed from the state correlation processing unit 16. The point which is set as the part 16B, and the point further provided with the prediction process part 32 and the test process part 33 differ.

  Next, the operation of the tracking device according to the fourth embodiment will be described with reference to the drawings.

  FIG. 12 is a flowchart showing the operation of the tracking device according to Embodiment 4 of the present invention. Steps 401-410 and 414 in FIG. 12 correspond to steps 101-110 and 112 in FIG. 2, respectively.

  In FIG. 12, the operation of the tracking device 1C according to the fourth embodiment of the present invention is compared with the operation of the first embodiment described above, instead of step 111, from step 411 (prediction processing) to step 413 (test processing). ) Is different. Therefore, the operation of the tracking device 1C according to the fourth embodiment will be described below with a focus on these differences.

First, by the operation up to step 410, the state correlation processing unit 16B outputs state vector smooth values x ₁ ^*** to x _n ^*** from the groups Γ ₁ to Γ _n . From this state vector smooth value x ₁ ^*** to x _n ^***, the state vector at the latest time t _N is estimated.

Next, in step 411, the prediction processing unit 32 sets the input state vector smooth values x ₁ ^*** to x _n ^{*** for} each group by one sample future time from the latest time according to a pre-estimated target motion. t to _{n + 1} forward extrapolation, _{x n,} by calculating the to ^{n + 1 ***} outputted from the predicted value _{x 1,} ^{n + 1 ***} state vector of each group.

Next, in step 412, the observation data management unit 12 acquires from the observation data storage unit 11 observation data z _{N + 1, 1} to z _{N + 1, m of} one sample future time t _{N + 1} from the latest time, and performs a verification processing unit 33. Output to.

Next, in step 413, the test processing unit 33 calculates the correspondence relationship between the state vector predicted values x _{j, N + 1} ^*** of the group Γ _j and the observed data z _{N + 1, k} at the time t _{N + 1} using the formula (3 ) And the distance d _{j, k} ″ based on a predetermined matrix E _{N + 1 as shown} in FIG. Then, the test processing unit 33 determines the state vector smooth value x _j ^{*** of the} group Γ _j where the observation data z _{N + 1, k} exists such that the distance d _{j, k} ″ is equal to or less than the predetermined threshold value d _G ′. And φ (Z, x _i ^*** ) are output to the wake establishment determination unit 22.

However, all groups Γ distance in _{j d j, k ''} is the threshold _{d G} 'and if the following become observation data _{z N + 1, k} does not exist, a distance _{d j, k'} 'is the threshold _{d G'} less If there are a plurality of groups in which observation data z _{N + 1, k} exists, state vector smooth values x _j ^*** and φ (Z, x _j ^{*) of the} group having the largest number of state vector estimation values belonging to the group ^{. **} ) is selected and output to the wake establishment determination unit 22.

As described above, according to the fourth embodiment, the state vector estimated from the observation data from time t ₁ to t _N is verified using the observation data at time t _{N + 1} to converge to the local maximum point. It is possible to determine and eliminate the estimated state vector value.

In the fourth embodiment, the distance between the state vector estimated value x _k ^** and the state vector predicted value x _{i, k} ^** is given by equation (1) in step 408 as in the first embodiment. Although described as a thing, it cannot be overemphasized that it can also give by Formula (2).

  1, 1A, 1B, 1C tracking device, 11 observation data storage unit, 12 observation data management unit, 13 selection time state candidate generation unit, 14 state space rough search unit, 15 state space detailed search unit, 16 state correlation processing unit, 16A state correlation processing unit, 16B state correlation processing unit, 17 prediction processing unit, 18 gate inside / outside determination unit, 19 grouping processing unit, 20 smoothing processing unit, 21 track determination unit, 22 track establishment determination unit, 23 reference time setting unit, 24 state vector estimated value storage unit, 25 latest time state vector estimated value storage unit, 26 latest time state correlation processing unit, 27 prediction processing unit, 28 gate inside / outside determination unit, 29 grouping processing unit, 30 smoothing processing unit, 31 track determination Part, 32 prediction processing part, 33 test processing part.

Claims (8)

An observation data storage unit that stores observation data input from the sensor at each observation time;
An observation data management unit that selects which observation data is used for estimation based on the wake establishment determination result among the observation data stored in the observation data storage unit;
A selection time state candidate generation unit that sets all observation times selected by the observation data management unit as reference times, and generates wake candidates at each reference time as state vector candidates;
Using the observation data selected by the observation data management unit, a likelihood ratio of the state vector candidates is calculated, and a state space rough search unit for searching for a state vector candidate having the maximum likelihood ratio at each reference time; ,
Using the observation data selected by the observation data management unit as an initial value, the state vector candidate having the maximum likelihood ratio at each reference time searched by the state space approximate search unit, and using the observation data selected by the observation data management unit, A state space detailed search unit for searching for a state vector estimated value having the maximum likelihood ratio;
A state correlation processing unit for determining the most probable state vector estimated value of the latest time from the correspondence relationship of the state vector estimated value having the maximum likelihood ratio at each reference time;
A tracking device comprising: a wake establishment determination unit that determines wake establishment from the most probable state vector estimation value of the latest time and outputs the wake establishment determination result. The state correlation processing unit
A prediction processing unit that extrapolates the state vector estimated value with the maximum likelihood ratio at each reference time forward according to a previously assumed target motion and calculates the state vector predicted value until reaching the latest time;
A gate internal / external determination unit that determines whether or not there is a correspondence relationship between the state vector estimated value at each reference time and the state vector predicted value calculated by the prediction processing unit;
About the state vector estimated value at each reference time, a grouping processing unit that collects the state vector estimated values determined to have a correspondence by the gate inside / outside determining unit as one group,
A smoothing processing unit that obtains a state vector smoothed value at the latest time of each group by smoothing the state vector estimated values grouped by the grouping processing unit;
For the state vector smooth value at the latest time of each group, using the observation data selected by the observation data management unit, calculate the likelihood ratio and select the state vector smooth value of the group with the maximum likelihood ratio. The tracking device according to claim 1, further comprising a track determination unit that determines a track. Of the observation time selected by the observation data management unit, further comprising a reference time setting unit for setting a predetermined observation time as the current reference time based on the track determination result from the track determination unit,
The selection time state candidate generation unit generates a wake candidate at the current reference time as a state vector candidate,
The state space rough search unit calculates the likelihood ratio of the state vector candidate using the observation data selected by the observation data management unit, and the state vector having the maximum likelihood ratio at the current reference time Search for candidates,
The state space detailed search unit uses the observation data selected by the observation data management unit using the state vector candidate having the maximum likelihood ratio at the current reference time searched by the state space approximate search unit as an initial value. To find a state vector estimate with the maximum likelihood ratio at the current reference time by iterative calculation,
The state correlation processing unit
A state vector estimated value storage unit that accumulates a state vector estimated value at the current reference time searched by the state space detailed search unit;
The prediction processing unit forwardly extrapolates the state vector estimated value group at the past reference time accumulated in the state vector estimated value storage unit to the current reference time according to a previously assumed target motion, and obtains a state vector predicted value. Calculate
The gate inside / outside determination unit determines whether or not there is a correspondence relationship between the state vector estimated value at the past and the current reference time and the state vector predicted value calculated by the prediction processing unit,
The grouping processing unit compiles the state vector estimation values determined to have a correspondence relationship by the gate internal / external determination unit as one group with respect to the state vector estimation values at the past and current reference times. 2. The tracking device according to 2. A latest time state vector estimated value storage unit for storing the most probable state vector estimated value of the latest time determined by the state correlation processing unit;
A latest time state correlation processing unit for determining the most probable state vector estimated value of the latest time from the correspondence between the past and latest time state vector estimated values accumulated in the latest time state vector estimated value storage unit;
The latest time state correlation processing unit
A prediction processing unit that extrapolates the state vector estimated value with the maximum likelihood ratio at the past and the latest time forward to the latest time according to a previously assumed target motion and calculates a state vector predicted value;
A gate internal / external determination unit that determines whether or not there is a correspondence relationship between the state vector estimated value at the past and the latest time and the state vector predicted value calculated by the prediction processing unit;
About the state vector estimated values at the past and latest times, a grouping processing unit that combines the state vector estimated values determined to have a correspondence by the gate internal / external determining unit as one group,
A smoothing processing unit that obtains a state vector smoothed value at the latest time of each group by smoothing the state vector estimated values grouped by the grouping processing unit;
For the state vector smooth value at the latest time of each group, using the observation data selected by the observation data management unit, calculate the likelihood ratio and select the state vector smooth value of the group with the maximum likelihood ratio. The tracking device according to claim 1, further comprising a track determination unit that determines a track. The state correlation processing unit
A first prediction processing unit that extrapolates the state vector estimated value with the maximum likelihood ratio at each reference time forward according to a pre-determined target motion and arrives at the latest time to calculate a state vector predicted value;
A gate internal / external determination unit that determines whether or not there is a correspondence relationship between the state vector estimated value at each reference time and the state vector predicted value calculated by the first prediction processing unit;
About the state vector estimated value at each reference time, a grouping processing unit that collects the state vector estimated values determined to have a correspondence by the gate inside / outside determining unit as one group,
A smoothing unit that obtains a state vector smoothed value at the latest time of each group by smoothing the state vector estimation values grouped by the grouping processing unit,
A state vector smoothed value at the latest time of each group obtained by the smoothing unit is extrapolated forward from the latest time to one sample future time according to a previously assumed target motion, and a state vector predicted value of each group is calculated. Two prediction processing units;
Observation data at one sample future time from the latest time is acquired from the observation data storage unit, and the observation data at one sample future time from the latest time and the state vector predicted value of each group are the most The tracking device according to claim 1, further comprising a test processing unit that determines a probable state vector estimation value. The gate inside / outside determination unit determines the presence / absence of the correspondence based on a magnitude relationship between a distance between the state vector estimated value and the state vector predicted value and a predetermined threshold value. The tracking device according to any one of claims 2 to 5. The state space approximate search unit and the state space detailed search unit extract a plurality of state vector estimation values at the respective reference times from those having a large likelihood ratio,
The tracking device according to claim 1, wherein the state correlation processing unit determines the most probable state vector estimated value from a correspondence relationship between a plurality of state vector estimated values at each reference time. The said test | inspection process part determines the most probable state vector estimated value based on the magnitude relationship of the distance between the said observation data and the said state vector prediction value, and the predetermined threshold value. Tracking device.

Images