WO2021019667A1 - Target estimation device and target estimation method - Google Patents

Target estimation device and target estimation method Download PDF

Info

Publication number
WO2021019667A1
WO2021019667A1 PCT/JP2019/029768 JP2019029768W WO2021019667A1 WO 2021019667 A1 WO2021019667 A1 WO 2021019667A1 JP 2019029768 W JP2019029768 W JP 2019029768W WO 2021019667 A1 WO2021019667 A1 WO 2021019667A1
Authority
WO
WIPO (PCT)
Prior art keywords
distribution
target
processing unit
targets
predicted value
Prior art date
Application number
PCT/JP2019/029768
Other languages
French (fr)
Japanese (ja)
Inventor
響介 小西
Original Assignee
三菱電機株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 三菱電機株式会社 filed Critical 三菱電機株式会社
Priority to JP2021536503A priority Critical patent/JP6945778B2/en
Priority to PCT/JP2019/029768 priority patent/WO2021019667A1/en
Publication of WO2021019667A1 publication Critical patent/WO2021019667A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/66—Radar-tracking systems; Analogous systems

Definitions

  • the present invention relates to a target estimation device and a target estimation method for estimating the number of targets and the state of targets.
  • Non-Patent Document 1 describes a CPHD filter (Cardinalized Probability Hypothesis Density filter).
  • the CPHD filter has a probability distribution (hereinafter referred to as a number distribution) representing the number of targets in the observation range and a probability hypothesis density distribution (hereinafter referred to as a number distribution) representing the density of the average number of targets in the state space representing the target state.
  • the number and state of the targets are estimated by sequentially updating the PHD distribution) based on the motion model and the observation model of each of the plurality of targets.
  • the estimated value of the number of targets is unstable under conditions where it is difficult to observe the targets.
  • the number distribution of the target has a large variance.
  • the number of targets having the maximum number distribution of targets is calculated as the estimated value of the number of targets. Therefore, when the variance of the number distribution is large, the estimated value of the number of targets calculated sequentially is time. The value becomes unstable with the passage of time.
  • the present invention solves the above problems, and an object of the present invention is to obtain an apparatus and a method capable of calculating an estimated value of a target state and an estimated value of a stable number of targets over time. ..
  • the target estimation device has a first distribution which is a probability distribution representing the number of targets in the observation range and a second distribution which represents the density of the average number of targets in the state space representing the state of the targets.
  • a device that estimates the number and state of targets based on the distribution of a prediction processing unit that calculates the predicted value of the first distribution and the predicted value of the second distribution, and the distribution within the specified range. Based on the predicted value of the first distribution or the distribution deformation part that deforms the first distribution so that it fits in, the predicted value of the first distribution, the predicted value of the second distribution, and the observation data of the motion specifications of the target. It is provided with an update processing unit that calculates the first distribution and the second distribution, and an estimation processing unit that sequentially estimates the number and states of targets based on the calculated first distribution and second distribution. ..
  • the first distribution is a probability distribution representing the number of targets in the observation range
  • the second distribution is a distribution representing the density of the average number of targets in the state space representing the state of the targets.
  • FIG. 3A is a block diagram showing a hardware configuration that realizes the function of the target estimation device according to the first embodiment
  • FIG. 3B is a block diagram that executes software that realizes the function of the target estimation device according to the first embodiment.
  • FIG. 1 is a block diagram showing a configuration of the target estimation device 1 according to the first embodiment.
  • the target estimation device 1 estimates the number and state of targets using the observation data A of the motion specifications of the target stored in the observation data storage unit 2, and stores the estimated value B of the number and states of targets in the estimated value storage unit.
  • the target state is, for example, the target position and velocity, and a vector representing the position and velocity of one target is called a “state vector”.
  • the observation data A is a time series of data representing the position of the target.
  • the observation data A is observed by, for example, a sensor or a radar device, and is sequentially stored in the observation data storage unit 2.
  • the target state is not limited to the target position and speed, and may be any motion specifications that can be observed as the target state.
  • the target position, velocity, and acceleration may be the target states.
  • the target state vector is a vector representing the target position, velocity, and acceleration.
  • the observation data A may be a time series of data representing the position and speed of the target.
  • the target estimation device 1 estimates the number and state of targets based on the number distribution C of targets and the PHD (Probability Hypothesis Density) distribution D of targets.
  • the target number distribution C is a first distribution representing the number of targets within the observation range of the target estimation device 1, and is a probability distribution of the number of targets within the observation range. Further, the number distribution C can be represented by a probability density function of the number of targets existing in the observation range at a certain time. However, in order to treat the number distribution C as data having a finite size, an approximation is used in which an upper limit value n max is set for the number of targets and the probability that the number of targets is equal to or more than the upper limit value n max is 0. As a result, the number distribution C is defined as a vector value of (n max +1) dimension.
  • the target estimation device 1 can estimate the number of targets for each observation time by calculating the target number distribution C for each observation time.
  • the target PHD distribution D is a second distribution representing the density of the average number of targets in the state space representing the target state.
  • the state space is the space of the target state vector.
  • the target state vector is a vector having a total of 6-dimensional elements consisting of a 3-dimensional position and a 3-dimensional velocity
  • the distribution of the average number of targets per unit volume in the 6-dimensional vector space is It is a PHD distribution D.
  • the value obtained by integrating the PHD distribution D over the entire space of the state vector corresponds to the average value of the number of targets existing regardless of the state, which corresponds to the average value of the number distribution.
  • the target estimation device 1 can estimate the target state vector for each observation time by calculating the target PHD distribution D for each observation time.
  • the target estimation device 1 calculates the estimated value of the number of targets and the state at each observation time based on the target number distribution C and the PHD distribution D recursively calculated inside the device.
  • the estimation process of the number of targets and the state at time k which is the observation time when the observation data A was acquired at the kth time, will be described.
  • k is a natural number of 2 or more.
  • the estimation is performed in the direction in which the time advances. For example, when the number and state of targets at time k are estimated, the number and state of targets at time k + 1 are estimated next.
  • the target estimation device 1 includes a prediction processing unit 11, a number distribution deformation unit 12, an update processing unit 13, and an estimation processing unit 14.
  • the prediction processing unit 11 calculates the predicted value Ca of the number distribution and the predicted value Da of the PHD distribution at time k based on the number distribution C and the PHD distribution D at time k-1 calculated by the update processing unit 13.
  • the predicted value Ca of the number distribution at time k calculated by the prediction processing unit 11 is output to the number distribution transforming unit 12, and the predicted value Da of the PHD distribution at time k is output to the update processing unit 13.
  • the number distribution transforming unit 12 is a distribution transforming unit that transforms the number distribution of the predicted value Ca at time k so that the variance of the number distribution falls within a designated range.
  • the predicted value Cab of the number distribution at the time k when the distribution is deformed by the number distribution deformation unit 12 is output to the update processing unit 13.
  • the update processing unit 13 determines the number distribution C and the PHD distribution D at time k based on the predicted value Cab of the number distribution at time k, the predicted value Da of the PHD distribution at time k, and the observation data A of the target state at time k. Is calculated.
  • the number distribution C and the PHD distribution D at time k calculated by the update processing unit 13 are output to the estimation processing unit 14 and further output to the prediction processing unit 11.
  • the prediction processing unit 11 determines the predicted value Ca of the number distribution at the next time k + 1 and the predicted value Da of the PHD distribution. calculate.
  • the estimation processing unit 14 estimates the number and state of targets at time k based on the number distribution C and PHD distribution D at time k calculated by the update processing unit 13.
  • the estimated value B of the number of targets and the state at time k calculated by the estimation processing unit 14 is stored in the estimated value storage unit 3.
  • each state of the plurality of targets is represented by the state vector x.
  • the state vector x is a six-dimensional vector representing the three-dimensional position and the three-dimensional velocity of the target. It is assumed that one observation data A at time k is represented by the observation vector z k .
  • the observation data A when the target is representative of the position of the three-dimensional detected, the observation vector z k is a three-dimensional vector. Since the number of observation data A acquired at time k is not always one, all observation data A acquired at time k are represented by a set Z k of observation vectors z k . If the target is not observed at time k, then Z k is the empty set. The number of elements included in the set Z k is expressed as
  • the number distribution C at time k is represented by p k (n), and the number distribution C at time k-1 is represented by p k-1 (n).
  • the predicted value of the number distribution at time k shall be expressed as pk
  • the PHD distribution D at time k is represented by v k (x), and the PHD distribution D at time k-1 is represented by v k-1 (x).
  • the predicted value of the PHD distribution at time k shall be expressed as v k
  • FIG. 2 is a flowchart showing the method according to the first embodiment, and shows a series of processes until the target estimation device 1 estimates the number and the state of the targets.
  • the prediction processing unit 11 executes prediction processing of the target number distribution and the PHD distribution (step ST1). For example, the prediction processing unit 11 uses the number distribution p k-1 (n) and the PHD distribution v k-1 (n) at the time k-1 calculated by the update processing unit 13 to use the following equation (3) and Based on the motion model according to the following equation (4), the predicted value of the number distribution at time k
  • k-1 [v, p] (j) in the following formula (3) is represented by the following formula (5).
  • p ⁇ and k (n) are parameters corresponding to the number distribution of targets appearing at time k
  • ⁇ k (x) is the target appearing at time k. It is a parameter corresponding to the PHD distribution, and is a parameter indicating the frequency of increase in the number of targets.
  • k-1 (x) represents the probability that the target of the state x remains without disappearing between the time k-1 and the time k. This is a parameter that indicates the frequency of decrease in the number of targets.
  • ⁇ ) in the above equation (4) represents the probability that the target transitions from the state ⁇ to the state x between the time k-1 and the time k, and represents the motion model of the target. It is a parameter.
  • C i j in the above equation (5) is a binomial coefficient representing the number of cases where the i number choose j number.
  • the integral for the state vector in the above equations (4) and (5) can be calculated by approximating the PHD distribution v k-1 (x) at time k-1 using the mixed Gaussian model or the sequential Monte Carlo method. .. Further, for the infinite series in the above equation (5), the maximum value of the number of targets is determined, and an approximation of pk
  • k-1 (n) 0 is used for n which is sufficiently larger than this maximum value. It becomes possible to calculate with.
  • the number distribution deformation unit 12 sets the shape parameter (step ST2).
  • the shape parameter is a parameter used to transform the number distribution.
  • the shape parameter prevents the shape of the number distribution from being excessively widened or narrowed, and keeps the variance of the number distribution after deformation within a specified range.
  • the shape parameter must be set so that the average value of the number distribution is the same before and after the deformation. This is because the average value of the number distribution indicates the average number of targets at time k and includes the information of the observation data A acquired in the past. That is, when the number distribution after deformation is p'k
  • the shape parameter depends on the number distribution p'k
  • r, ⁇ and ⁇ are shape parameters, and all of them are real numbers.
  • the range of values that the shape parameters r, ⁇ , and ⁇ can take is the range shown in the following equations (8) to (10).
  • k-1 (n) represented by the above formula (7) is represented by the following formula (12). Further, the variance ⁇ 2 of the number distribution p'k
  • the number distribution deformation unit 12 searches for shape parameters r, ⁇ , and ⁇ that satisfy the following equations (14) and (15).
  • ⁇ 2 L is a parameter that specifies the lower limit of the variance
  • ⁇ 2 H is a parameter that specifies the upper limit of the variance.
  • An optimization algorithm such as the steepest descent method can be used to search for shape parameters. That is, the specified range for the variance ⁇ 2 of the number distribution C after deformation is determined by the following equation (15) determined by the lower limit value ⁇ 2 L and the upper limit value ⁇ 2 H using the shape parameters r, ⁇ and ⁇ . It is the range shown.
  • the formula for expressing the shape of the number distribution after deformation is not limited to the above formula (7), and the number distribution deformation unit 12 can use a shape represented by another formula.
  • the shape of the number distribution after deformation may be represented by the double Poisson distribution represented by the following equation (16).
  • c is a normalized constant and e is the base of the natural logarithm.
  • the shape parameters are ⁇ and ⁇ .
  • the designated range regarding the variance of the number distribution C after deformation is the range determined by the shape parameters ⁇ and ⁇ .
  • the number distribution transforming unit 12 transforms the shape of the number distribution, which is a predicted value at time k, based on the searched shape parameter (step ST3).
  • the number distribution transformation unit 12 uses the shape parameters r, ⁇ and ⁇ , and according to the above equation (7), at the predicted value at time k.
  • k-1 (n) after a certain transformation is calculated.
  • the update processing unit 13 executes an update process for calculating the target number distribution and the PHD distribution at time k (step ST4).
  • the update processing unit 13 has a predicted value p'k
  • using a k-1 (x) the following equation (17) and based on the observation model according to (18), the number distribution p k at time k (n) and PHD distribution v k (n) Is calculated.
  • the following equation (17) and Upsilon u k in (18) [v, Z] (n) is expressed by the following equation (19).
  • u is 0 or 1.
  • p K and k (x) represent the number distribution of erroneously detected targets
  • ⁇ k (x) is the PHD distribution of erroneously detected targets.
  • p D, k (x) represent the probability of observing a target in the state x at time k, and is also a parameter of the observation model regarding the number of targets.
  • x) represents the likelihood of the observation vector z with respect to the target of the state x at time k, and is a parameter of the observation model regarding the target state.
  • Z k ⁇ ⁇ z ⁇ is a set of differences obtained by removing the set ⁇ z ⁇ from the set Z k .
  • ⁇ j (A) is a basic symmetric equation of order j with respect to each element of the set A, and is represented by the following equation (20).
  • the set ⁇ k (v, Z) is represented by the following formula (21), and P n j is represented by the following formula (22).
  • the integral for the state vector in the above equations (19) and (20) is calculated by approximating the predicted value v k
  • the estimation processing unit 14 executes estimation processing of the number of targets and the state at time k (step ST5). For example, the estimation processing unit 14 uses the number distribution of the target at time k calculated by the update processing unit 13 p k (n) and PHD distribution v k (x), according to the following equation (23), the number of target The estimated value n k of is calculated. Further, the estimation processing unit 14 gives a state vector that gives the upper nk peaks having the largest extreme values among the peaks of the PHD distribution v k (x) when the estimated value nk of the number of targets is 1 or more. Is obtained, and this state vector is used as an estimated value of the target state.
  • the estimation processing unit 14 approximates the PHD distribution v k (x) by a mixed Gaussian model or a sequential Monte Carlo method, and gives the upper nk peaks having a large extremum from the approximated PHD distribution v k (x). Find the state vector.
  • the process of estimating the number of targets and the state at time k based on the target number distribution C and PHD distribution D at time k-1 and the observation data A of the target state at time k does not necessarily flow toward the future, but may flow toward the past.
  • the target estimation device 1 determines the number and states of targets at time k based on the target number distribution C and PHD distribution D at time k + 1 and the observation data A of the target state at time k, which is the next time. You may estimate.
  • the parameter indicating the fluctuation of the number of targets and the transition of the target state in the prediction process is used as the parameter of the model representing the transition from time k + 1 to time k.
  • the target estimation device 1 includes a processing circuit that executes the processes from step ST1 to step ST5 in FIG.
  • the processing circuit may be dedicated hardware or a CPU (Central Processing Unit) that executes a program stored in the memory.
  • FIG. 3A is a block diagram showing a hardware configuration that realizes the function of the target estimation device 1
  • FIG. 3B is a block diagram showing a hardware configuration that executes software that realizes the function of the target estimation device 1.
  • the input interface 100 is an interface that relays the observation data A of the target state input from the observation data storage unit 2 to the target estimation device 1.
  • the output interface 101 is an interface that relays the estimated value B of the number of targets and the state output from the target estimation device 1 to the estimated value storage unit 3.
  • the input interface 100 and the output interface 101 are, for example, DVI (registered trademark) (Digital Visual Interface), HDMI (registered trademark) (High-Definition Multimedia Interface), USB (Universal Serial Bus), Ethernet (registered trademark), or Ethernet (registered trademark). It may be a (Control Area Network) bus.
  • the processing circuit 102 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, or an ASIC (Application Specific Integrated Circuitd). Circuit), FPGA (Field-Programmable Gate Array), or a combination thereof is applicable.
  • the functions of the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14 in the target estimation device 1 may be realized by separate processing circuits, or these functions may be combined into one process. It may be realized by a circuit.
  • the functions of the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14 in the target estimation device 1 are software, firmware, or software and firmware. It is realized by the combination with.
  • the software or firmware is described as a program and stored in the memory 104.
  • the processor 103 realizes the functions of the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14 in the target estimation device 1 by reading and executing the program stored in the memory 104.
  • the target estimation device 1 includes a memory 104 for storing a program in which the processes of steps ST1 to ST5 in the flowchart shown in FIG. 2 are executed as a result when executed by the processor 103.
  • These programs cause a computer to execute the procedures or methods of the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14.
  • the memory 104 may be a computer-readable storage medium in which a program for causing the computer to function as the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14 is stored.
  • the memory 104 is, for example, a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory), an EEPROM (Electrically-volatile) semiconductor, an EPROM (Electrically-volatile), or the like.
  • the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14 in the target estimation device 1 are realized by dedicated hardware, and some of them are realized by software or firmware.
  • the prediction processing unit 11, the update processing unit 13, and the estimation processing unit 14 realize the functions by the processing circuit 102, which is dedicated hardware, and the number distribution transformation unit 12 is a program in which the processor 103 is stored in the memory 104. The function is realized by reading and executing. In this way, the processing circuit can realize the above functions by hardware, software, firmware or a combination thereof.
  • the target estimation device 1 includes a prediction processing unit 11, a number distribution deformation unit 12, an update processing unit 13, and an estimation processing unit 14.
  • the estimation processing unit 14 estimates the number and state of the target based on the target number distribution C and the PHD distribution D at time k
  • the number distribution deformation unit 12 so that the variance of the distribution falls within the specified range.
  • the target estimation device 1 can calculate an estimated value of the target state quantity and an estimated value of the number of targets that are stable over time.
  • Embodiment 2 In the first embodiment, a process of deforming the shape of the number distribution of the predicted values is performed so that the target number distribution has a desired spread.
  • the shape of the number distribution used in the number distribution prediction process is modified. Also in the second embodiment, since the target number distribution is recursively used in the prediction process and the update process, the same effect as that of the first embodiment can be obtained.
  • the parameters representing the fluctuation model of the number of targets used in the prediction processing by the prediction processing unit 11 in the first embodiment are, for example, the parameters p ⁇ , which represent the increase frequency of the targets in the above equations (3) and (4) . k (n) and ⁇ k (x). The values of these parameters are reflected only in the mean value of the number distribution, not in the shape of the number distribution.
  • the prediction processing unit 11A calculates the predicted value of the number distribution using the number distribution deformed by the number distribution deformation unit 12A, and inputs it to the update processing unit 13.
  • the parameters representing the variation model of the number of targets can be more strongly reflected in the estimation result of the number of targets and the state.
  • FIG. 4 is a block diagram showing the configuration of the target estimation device 1A according to the second embodiment.
  • the same components as those in FIG. 1 are designated by the same reference numerals, and the description thereof will be omitted.
  • the target estimation device 1A calculates the number of targets and the estimated value B of the state by using the observation data A of the target state stored in the observation data storage unit 2, and stores the estimated value. Store in part 3.
  • the target estimation device 1A includes a prediction processing unit 11A, a number distribution deformation unit 12A, an update processing unit 13, and an estimation processing unit 14.
  • the prediction processing unit 11A is based on the target number distribution Cb at the time k-1 deformed by the number distribution transforming unit 12A and the target PHD distribution D at the time k-1 calculated by the update processing unit 13.
  • the predicted value Cab of the number distribution in k and the predicted value Da of the PHD distribution are calculated.
  • the predicted value Cab of the number distribution and the predicted value Da of the PHD distribution at time k calculated by the prediction processing unit 11A are output to the update processing unit 13.
  • the update processing unit 13 determines the number distribution C and the PHD distribution D at time k based on the predicted value Cab of the number distribution at time k, the predicted value Da of the PHD distribution at time k, and the observation data A of the target state at time k. Is calculated.
  • the number distribution deformation unit 12A is a distribution deformation unit that transforms the number distribution C at time k-1 calculated by the update processing unit 13 so that the variance of the number distribution C falls within a designated range.
  • the number distribution Cb at time k-1 in which the distribution is deformed by the number distribution transforming unit 12A is output to the prediction processing unit 11A.
  • FIG. 5 is a flowchart showing the method according to the second embodiment, and shows a series of processes until the target estimation device 1A estimates the number and the state of the targets.
  • the number distribution deformation unit 12A sets the shape parameter (step ST1a).
  • the shape parameter is a parameter used to transform the number distribution, as in the first embodiment.
  • the shape parameter prevents the shape of the number distribution from being excessively widened or narrowed, and keeps the variance of the number distribution after deformation within a specified range.
  • the shape parameter must be set so that the average value of the number distribution is the same before and after the deformation. This is because the average value of the number distribution indicates the average number of targets at time k and includes the information of the observation data A acquired in the past.
  • the number distribution deformation unit 12A searches for shape parameters r, ⁇ , and ⁇ that satisfy the above equations (14) and (15).
  • An optimization algorithm such as the steepest descent method can be used to search for shape parameters. That is, the specified range for the variance ⁇ 2 of the number distribution C after deformation is determined by the above equation (15) determined by the lower limit value ⁇ 2 L and the upper limit value ⁇ 2 H using the shape parameters r, ⁇ and ⁇ . It is the range shown.
  • the formula for expressing the shape of the number distribution after deformation is not limited to the above formula (7), and the number distribution deformation unit 12A can use a shape represented by another formula.
  • the shape of the number distribution after deformation may be represented by the double Poisson distribution represented by the above equation (16).
  • c is a normalized constant and e is the base of the natural logarithm.
  • the shape parameters are ⁇ and ⁇ .
  • the designated range regarding the variance of the number distribution C after deformation is the range determined by the shape parameters ⁇ and ⁇ .
  • the number distribution deformation unit 12A deforms the shape of the number distribution at time k-1 based on the searched shape parameter (step ST2a). For example, when the number distribution after deformation is represented by the above formula (7), the number distribution deformation unit 12A uses the shape parameters r, ⁇ and ⁇ and is deformed at time k-1 according to the above formula (7). The latter number distribution Cb is calculated.
  • the prediction processing unit 11A executes the prediction processing of the target number distribution and the PHD distribution (step ST3a). For example, the prediction processing unit 11A uses the number distribution at time k-1 deformed by the number distribution transforming unit 12A and the PHD distribution at time k-1 calculated by the update processing unit 13 to use the above equation (3). ) And the predicted value of the number distribution and the predicted value of the PHD distribution at time k are calculated based on the motion model according to the above equation (4).
  • the update processing unit 13 executes an update process for calculating the target number distribution and PHD distribution at time k (step ST4a). For example, the update processing unit 13 uses the predicted value of the number distribution and the predicted value of the PHD distribution at time k calculated by the prediction processing unit 11A, and the time is based on the observation model according to the above equations (17) and (18). Calculate the target number distribution and PHD distribution in k.
  • the estimation processing unit 14 executes estimation processing of the number of targets and the state at time k (step ST5a). For example, the estimation processing unit 14 calculates an estimated value n k of the number of targets according to the above equation (23) using the target number distribution and the PHD distribution at time k calculated by the update processing unit 13. When the estimated value n k of the number of targets is 1 or more, the estimation processing unit 14 obtains a state vector that gives the upper nk peaks having a large extreme value among the peaks of the PHD distribution, and obtains this state vector. It is an estimate of the target state. For example, the estimation processing unit 14 approximates the PHD distribution by a mixed Gaussian model or a sequential Monte Carlo method, and obtains a state vector that gives the upper nk peaks having large extreme values from the approximated PHD distribution.
  • the process of estimating the number of targets and the state at time k is shown based on the target number distribution C and PHD distribution D at time k-1 and the observation data A of the target state at time k.
  • the time for estimating the number and state of targets does not necessarily flow toward the future, but may flow toward the past.
  • the target estimation device 1A may estimate the number and state of targets at time k based on the number distribution C and PHD distribution D of targets at time k + 1 and the observation data A of the state of the target at time k. ..
  • the parameter indicating the change in the number of targets and the transition of the target state in the prediction process is used as the parameter related to the model indicating the transition from time k + 1 to time k.
  • the target estimation device 1A includes a processing circuit for executing the processing of steps ST1a to ST5a shown in FIG.
  • the processing circuit may be the processing circuit 102 of the dedicated hardware shown in FIG. 3A, or the processor 103 that executes the program stored in the memory 104 shown in FIG. 3B.
  • the target estimation device 1A includes a prediction processing unit 11A, a number distribution deformation unit 12A, an update processing unit 13, and an estimation processing unit 14.
  • the estimation processing unit 14 estimates the number and state of the target based on the target number distribution C and the PHD distribution D at time k
  • the number distribution deformation unit 12A is set so that the distribution variance is within the specified range.
  • the number distribution C at time k-1 is transformed.
  • the prediction processing unit 11A is based on the target number distribution Cb at time k-1 deformed by the number distribution transforming unit 12A and the target PHD distribution D at time k-1 calculated by the update processing unit 13.
  • the predicted value Cab of the number distribution at time k and the predicted value Da of the PHD distribution are calculated.
  • the parameters representing the fluctuation model of the number of targets can be more strongly reflected in the estimation result of the number of targets and the state.
  • the existing CPHD filter can be used. In comparison, the frequency of deterioration in the number of targets and the estimation accuracy of the state can be reduced.
  • the present invention is not limited to the above-described embodiment, and within the scope of the present invention, any combination of the embodiments or any component of the embodiment may be modified or the embodiment. Any component can be omitted in each of the above.
  • the device according to the present invention can estimate the number and state of targets, for example, targeting vehicles, ships, aircraft, robots, people or bicycles.
  • 1,1A target estimation device 2 observation data storage unit, 3 estimation value storage unit, 11,11A prediction processing unit, 12,12A number distribution deformation unit, 13 update processing unit, 14 estimation processing unit, 100 input interface, 101 output Interface, 102 processing circuit, 103 processor, 104 memory.

Landscapes

  • Engineering & Computer Science (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Radar Systems Or Details Thereof (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A target estimation device (1) is provided with: a prediction processing unit (11) that calculates a predictive value of a number distribution and a predictive value of a PHD distribution; a number distribution deformation unit (12) that deforms the predictive value of the number distribution such that the dispersion of distribution falls within a designated range; an update processing unit (13) that, on the basis of the predictive value of the deformed number distribution, the predictive value of the PHD distribution, and target observation data, calculates the number distribution and the PHD distribution; an estimation processing unit (14) that, on the basis of the number distribution and the PHD distribution, sequentially estimates a target number and a target state.

Description

目標推定装置および目標推定方法Target estimation device and target estimation method
　本発明は、目標の個数および目標の状態を推定する目標推定装置および目標推定方法に関する。 The present invention relates to a target estimation device and a target estimation method for estimating the number of targets and the state of targets.
　従来から、車両、船舶、航空機、ロボットまたは人を目標として、センサによって観測された目標の運動諸元（例えば、目標の位置）の観測データの時系列に基づいて、目標の状態（例えば、目標の位置、速度および加速度）と目標の個数を逐次的に推定する技術が知られている。例えば、非特許文献１には、ＣＰＨＤフィルタ（Ｃａｒｄｉｎａｌｉｚｅｄ　Ｐｒｏｂａｂｉｌｉｔｙ　Ｈｙｐｏｔｈｅｓｉｓ　Ｄｅｎｓｉｔｙ　ｆｉｌｔｅｒ）が記載されている。ＣＰＨＤフィルタは、観測範囲内における目標の個数を表す確率分布（以下、個数分布と記載する）と、目標の状態を表す状態空間内における目標の平均個数の密度を表す確率仮説密度分布（以下、ＰＨＤ分布と記載する）を、複数の目標のそれぞれの運動モデルおよび観測モデルに基づいて逐次更新することによって、目標の個数および状態を推定するものである。 Traditionally, targeting a vehicle, ship, aircraft, robot, or person, the state of the target (eg, target) is based on the time series of observation data of the motion specifications (eg, target position) of the target observed by the sensor. (Position, speed and acceleration) and the number of targets are known in sequence. For example, Non-Patent Document 1 describes a CPHD filter (Cardinalized Probability Hypothesis Density filter). The CPHD filter has a probability distribution (hereinafter referred to as a number distribution) representing the number of targets in the observation range and a probability hypothesis density distribution (hereinafter referred to as a number distribution) representing the density of the average number of targets in the state space representing the target state. The number and state of the targets are estimated by sequentially updating the PHD distribution) based on the motion model and the observation model of each of the plurality of targets.
　非特許文献１に記載されたＣＰＨＤフィルタでは、目標の個数分布およびＰＨＤ分布に基づいて目標の個数および状態を推定する際、目標の観測が困難な条件において、目標の個数の推定値が不安定になるという課題があった。ＣＰＨＤフィルタでは、目標の個数の推定値を逐次的に算出する際に、目標の個数分布の形状に前提が設けられていない。このため、目標の観測が困難な条件、すなわち、目標の観測データの誤検出および欠落が頻発する場合、目標の個数分布は、分散が大きくなる。ＣＰＨＤフィルタでは、目標の個数の推定値として目標の個数分布が最大となる目標の個数を算出するので、個数分布の分散が大きい場合、逐次的に算出される目標の個数の推定値は、時間経過に対して不安定な値になる。 In the CPHD filter described in Non-Patent Document 1, when estimating the number and state of targets based on the number distribution of targets and the PHD distribution, the estimated value of the number of targets is unstable under conditions where it is difficult to observe the targets. There was a problem of becoming. In the CPHD filter, when the estimated value of the target number is sequentially calculated, no premise is provided for the shape of the target number distribution. Therefore, under conditions where it is difficult to observe the target, that is, when erroneous detection and omission of the target observation data occur frequently, the number distribution of the target has a large variance. In the CPHD filter, the number of targets having the maximum number distribution of targets is calculated as the estimated value of the number of targets. Therefore, when the variance of the number distribution is large, the estimated value of the number of targets calculated sequentially is time. The value becomes unstable with the passage of time.
　本発明は上記課題を解決するものであって、目標の状態の推定値と、時間経過に対して安定した目標の個数の推定値を算出することができる装置および方法を得ることを目的とする。 The present invention solves the above problems, and an object of the present invention is to obtain an apparatus and a method capable of calculating an estimated value of a target state and an estimated value of a stable number of targets over time. ..
　本発明に係る目標推定装置は、観測範囲内における目標の個数を表す確率分布である第１の分布と、目標の状態を表す状態空間内における目標の平均個数の密度を表す分布である第２の分布に基づいて、目標の個数および状態を推定する装置であって、第１の分布の予測値および第２の分布の予測値を算出する予測処理部と、分布の分散が指定した範囲内に収まるように第１の分布の予測値または第１の分布を変形する分布変形部と、第１の分布の予測値、第２の分布の予測値および目標の運動諸元の観測データに基づいて、第１の分布および第２の分布を算出する更新処理部と、算出された第１の分布および第２の分布に基づいて目標の個数および状態を逐次的に推定する推定処理部を備える。 The target estimation device according to the present invention has a first distribution which is a probability distribution representing the number of targets in the observation range and a second distribution which represents the density of the average number of targets in the state space representing the state of the targets. A device that estimates the number and state of targets based on the distribution of, a prediction processing unit that calculates the predicted value of the first distribution and the predicted value of the second distribution, and the distribution within the specified range. Based on the predicted value of the first distribution or the distribution deformation part that deforms the first distribution so that it fits in, the predicted value of the first distribution, the predicted value of the second distribution, and the observation data of the motion specifications of the target. It is provided with an update processing unit that calculates the first distribution and the second distribution, and an estimation processing unit that sequentially estimates the number and states of targets based on the calculated first distribution and second distribution. ..
　本発明によれば、観測範囲内における目標の個数を表す確率分布である第１の分布と、目標の状態を表す状態空間内における目標の平均個数の密度を表す分布である第２の分布に基づいて目標の個数および状態を推定する際、分布の分散が指定した範囲内に収まるように第１の分布予測値または第１の分布を変形する。これにより、目標の観測データの誤検出および欠落が頻発する場合であっても、目標の状態量の推定値と、時間経過に対して安定した目標の個数の推定値を算出することができる。 According to the present invention, the first distribution is a probability distribution representing the number of targets in the observation range, and the second distribution is a distribution representing the density of the average number of targets in the state space representing the state of the targets. When estimating the number and state of targets based on it, the first distribution prediction value or the first distribution is modified so that the variance of the distribution falls within the specified range. As a result, even when erroneous detection and omission of the target observation data occur frequently, it is possible to calculate the estimated value of the state quantity of the target and the estimated value of the number of targets stable over time.
実施の形態１に係る目標推定装置の構成を示すブロック図である。It is a block diagram which shows the structure of the target estimation apparatus which concerns on Embodiment 1. FIG. 実施の形態１に係る目標推定方法を示すフローチャートである。It is a flowchart which shows the target estimation method which concerns on Embodiment 1. FIG. 図３Ａは、実施の形態１に係る目標推定装置の機能を実現するハードウェア構成を示すブロック図であり、図３Ｂは、実施の形態１に係る目標推定装置の機能を実現するソフトウェアを実行するハードウェア構成を示すブロック図である。FIG. 3A is a block diagram showing a hardware configuration that realizes the function of the target estimation device according to the first embodiment, and FIG. 3B is a block diagram that executes software that realizes the function of the target estimation device according to the first embodiment. It is a block diagram which shows the hardware configuration. 実施の形態２に係る目標推定装置の構成を示すブロック図である。It is a block diagram which shows the structure of the target estimation apparatus which concerns on Embodiment 2. FIG. 実施の形態２に係る目標推定方法を示すフローチャートである。It is a flowchart which shows the target estimation method which concerns on Embodiment 2.
実施の形態１．
　図１は、実施の形態１に係る目標推定装置１の構成を示すブロック図である。目標推定装置１は、観測データ記憶部２に記憶された目標の運動諸元の観測データＡを用いて目標の個数および状態を推定し、目標の個数および状態の推定値Ｂを推定値記憶部３に記憶する。目標の状態は、例えば、目標の位置および速度であり、１個の目標の位置および速度を表すベクトルを、“状態ベクトル”と呼ぶ。観測データＡは、目標の位置を表すデータの時系列であるものとする。観測データＡは、例えば、センサまたはレーダ装置によって観測されて観測データ記憶部２に逐次記憶される。 Embodiment 1.
FIG. 1 is a block diagram showing a configuration of the target estimation device 1 according to the first embodiment. The target estimation device 1 estimates the number and state of targets using the observation data A of the motion specifications of the target stored in the observation data storage unit 2, and stores the estimated value B of the number and states of targets in the estimated value storage unit. Store in 3. The target state is, for example, the target position and velocity, and a vector representing the position and velocity of one target is called a “state vector”. It is assumed that the observation data A is a time series of data representing the position of the target. The observation data A is observed by, for example, a sensor or a radar device, and is sequentially stored in the observation data storage unit 2.
　なお、実施の形態１において、目標の状態は、目標の位置および速度に限定されるものではなく、目標の状態として観測可能な運動諸元であればよい。例えば、目標の位置、速度および加速度を目標の状態としてもよい。この場合、目標の状態ベクトルは、目標の位置、速度および加速度を表すベクトルとなる。また、観測データＡは、目標の位置および速度を表すデータの時系列であってもよい。 Note that, in the first embodiment, the target state is not limited to the target position and speed, and may be any motion specifications that can be observed as the target state. For example, the target position, velocity, and acceleration may be the target states. In this case, the target state vector is a vector representing the target position, velocity, and acceleration. Further, the observation data A may be a time series of data representing the position and speed of the target.
　目標推定装置１は、目標の個数分布Ｃと、目標のＰＨＤ（Ｐｒｏｂａｂｉｌｉｔｙ　Ｈｙｐｏｔｈｅｓｉｓ　Ｄｅｎｓｉｔｙ）分布Ｄに基づいて、目標の個数および状態を推定する。目標の個数分布Ｃは、目標推定装置１の観測範囲内における目標の個数を表す第１の分布であり、観測範囲内における目標の個数の確率分布である。また、個数分布Ｃは、ある時刻で観測範囲に存在する目標の個数の確率密度関数で表すことができる。ただし、個数分布Ｃを有限な大きさのデータとして扱うために、目標の個数に上限値ｎｍａｘが定められ、目標の個数が上限値ｎｍａｘ以上である確率を０とする近似が用いられる。これにより、個数分布Ｃは、（ｎｍａｘ＋１）次元のベクトル値として定義される。目標推定装置１は、目標の個数分布Ｃを観測時刻ごとに算出することによって観測時刻ごとの目標の個数を推定することができる。 The target estimation device 1 estimates the number and state of targets based on the number distribution C of targets and the PHD (Probability Hypothesis Density) distribution D of targets. The target number distribution C is a first distribution representing the number of targets within the observation range of the target estimation device 1, and is a probability distribution of the number of targets within the observation range. Further, the number distribution C can be represented by a probability density function of the number of targets existing in the observation range at a certain time. However, in order to treat the number distribution C as data having a finite size, an approximation is used in which an upper limit value n max is set for the number of targets and the probability that the number of targets is equal to or more than the upper limit value n max is 0. As a result, the number distribution C is defined as a vector value of (n max +1) dimension. The target estimation device 1 can estimate the number of targets for each observation time by calculating the target number distribution C for each observation time.
　目標のＰＨＤ分布Ｄは、目標の状態を表す状態空間内における目標の平均個数の密度を表す第２の分布である。状態空間は、目標の状態ベクトルの空間である。例えば、目標の状態ベクトルが、３次元の位置と３次元の速度からなる合計６次元の要素を有したベクトルである場合、６次元ベクトル空間内における単位体積当たりの目標の平均個数の分布が、ＰＨＤ分布Ｄである。ＰＨＤ分布Ｄが状態ベクトルの空間全域で積分された値は、状態を問わず存在する目標の個数の平均値に相当し、これは、個数分布の平均値と一致する。目標推定装置１は、目標のＰＨＤ分布Ｄを観測時刻ごとに算出することで、観測時刻ごとの目標の状態ベクトルを推定することができる。 The target PHD distribution D is a second distribution representing the density of the average number of targets in the state space representing the target state. The state space is the space of the target state vector. For example, when the target state vector is a vector having a total of 6-dimensional elements consisting of a 3-dimensional position and a 3-dimensional velocity, the distribution of the average number of targets per unit volume in the 6-dimensional vector space is It is a PHD distribution D. The value obtained by integrating the PHD distribution D over the entire space of the state vector corresponds to the average value of the number of targets existing regardless of the state, which corresponds to the average value of the number distribution. The target estimation device 1 can estimate the target state vector for each observation time by calculating the target PHD distribution D for each observation time.
　目標推定装置１は、装置内部で再帰的に算出した目標の個数分布ＣとＰＨＤ分布Ｄとに基づいて、各観測時刻における目標の個数および状態の推定値を算出する。以下の説明では、ｋ回目に観測データＡが取得された観測時刻である時刻ｋにおける目標の個数および状態の推定処理について説明する。なお、ｋは２以上の自然数である。また、目標の個数および状態の推定処理では、時刻が進む方向に推定が実施されるものとする。例えば、時刻ｋにおける目標の個数および状態の推定が実施された場合、次は、時刻ｋ＋１における目標の個数および状態の推定が実施される。 The target estimation device 1 calculates the estimated value of the number of targets and the state at each observation time based on the target number distribution C and the PHD distribution D recursively calculated inside the device. In the following description, the estimation process of the number of targets and the state at time k, which is the observation time when the observation data A was acquired at the kth time, will be described. In addition, k is a natural number of 2 or more. Further, in the target number and state estimation process, the estimation is performed in the direction in which the time advances. For example, when the number and state of targets at time k are estimated, the number and state of targets at time k + 1 are estimated next.
　目標推定装置１は、図１に示すように、予測処理部１１、個数分布変形部１２、更新処理部１３および推定処理部１４を備えている。予測処理部１１は、更新処理部１３によって算出された時刻ｋ－１における個数分布ＣおよびＰＨＤ分布Ｄに基づいて、時刻ｋにおける個数分布の予測値ＣａおよびＰＨＤ分布の予測値Ｄａを算出する。予測処理部１１によって算出された時刻ｋにおける個数分布の予測値Ｃａは、個数分布変形部１２に出力され、時刻ｋにおけるＰＨＤ分布の予測値Ｄａは、更新処理部１３に出力される。 As shown in FIG. 1, the target estimation device 1 includes a prediction processing unit 11, a number distribution deformation unit 12, an update processing unit 13, and an estimation processing unit 14. The prediction processing unit 11 calculates the predicted value Ca of the number distribution and the predicted value Da of the PHD distribution at time k based on the number distribution C and the PHD distribution D at time k-1 calculated by the update processing unit 13. The predicted value Ca of the number distribution at time k calculated by the prediction processing unit 11 is output to the number distribution transforming unit 12, and the predicted value Da of the PHD distribution at time k is output to the update processing unit 13.
　個数分布変形部１２は、時刻ｋにおける予測値Ｃａの個数分布を、当該個数分布の分散が指定した範囲内に収まるように変形する分布変形部である。個数分布変形部１２によって分布が変形された時刻ｋにおける個数分布の予測値Ｃａｂは、更新処理部１３に出力される。更新処理部１３は、時刻ｋにおける個数分布の予測値Ｃａｂ、時刻ｋにおけるＰＨＤ分布の予測値Ｄａおよび時刻ｋにおける目標の状態の観測データＡに基づいて、時刻ｋにおける個数分布ＣおよびＰＨＤ分布Ｄを算出する。 The number distribution transforming unit 12 is a distribution transforming unit that transforms the number distribution of the predicted value Ca at time k so that the variance of the number distribution falls within a designated range. The predicted value Cab of the number distribution at the time k when the distribution is deformed by the number distribution deformation unit 12 is output to the update processing unit 13. The update processing unit 13 determines the number distribution C and the PHD distribution D at time k based on the predicted value Cab of the number distribution at time k, the predicted value Da of the PHD distribution at time k, and the observation data A of the target state at time k. Is calculated.
　更新処理部１３によって算出された時刻ｋにおける個数分布ＣおよびＰＨＤ分布Ｄは、推定処理部１４に出力され、さらに、予測処理部１１に出力される。予測処理部１１は、更新処理部１３によって算出された時刻ｋにおける個数分布ＣおよびＰＨＤ分布Ｄに基づいて、次の時刻である時刻ｋ＋１の個数分布の予測値ＣａおよびＰＨＤ分布の予測値Ｄａを算出する。推定処理部１４は、更新処理部１３によって算出された時刻ｋにおける個数分布ＣおよびＰＨＤ分布Ｄに基づいて時刻ｋにおける目標の個数および状態を推定する。推定処理部１４によって算出された時刻ｋにおける目標の個数および状態の推定値Ｂは、推定値記憶部３に記憶される。 The number distribution C and the PHD distribution D at time k calculated by the update processing unit 13 are output to the estimation processing unit 14 and further output to the prediction processing unit 11. Based on the number distribution C and PHD distribution D at time k calculated by the update processing unit 13, the prediction processing unit 11 determines the predicted value Ca of the number distribution at the next time k + 1 and the predicted value Da of the PHD distribution. calculate. The estimation processing unit 14 estimates the number and state of targets at time k based on the number distribution C and PHD distribution D at time k calculated by the update processing unit 13. The estimated value B of the number of targets and the state at time k calculated by the estimation processing unit 14 is stored in the estimated value storage unit 3.
　なお、以下の説明において、複数の目標のそれぞれの状態を状態ベクトルｘで表すものとする。例えば、目標の状態を、３次元空間内の目標の位置および速度とする場合、状態ベクトルｘは、目標の３次元位置と３次元速度とを表す６次元ベクトルとなる。時刻ｋにおける１つの観測データＡを観測ベクトルｚｋで表すものとする。 In the following description, it is assumed that each state of the plurality of targets is represented by the state vector x. For example, when the target state is the target position and velocity in the three-dimensional space, the state vector x is a six-dimensional vector representing the three-dimensional position and the three-dimensional velocity of the target. It is assumed that one observation data A at time k is represented by the observation vector z k .
　例えば、観測データＡが、目標が検出された３次元の位置を表す場合に、観測ベクトルｚｋは、３次元ベクトルとなる。時刻ｋにおいて取得される観測データＡは、常に１つとは限らないため、時刻ｋにおいて取得された全ての観測データＡを、観測ベクトルｚｋの集合Ｚｋで表す。時刻ｋで目標が観測されなかった場合、Ｚｋは空集合となる。集合Ｚｋに含まれる要素の個数を｜Ｚｋ｜と表記する。 For example, the observation data A, when the target is representative of the position of the three-dimensional detected, the observation vector z k is a three-dimensional vector. Since the number of observation data A acquired at time k is not always one, all observation data A acquired at time k are represented by a set Z k of observation vectors z k . If the target is not observed at time k, then Z k is the empty set. The number of elements included in the set Z k is expressed as | Z k |.
　また、時刻ｋにおける個数分布Ｃをｐｋ（ｎ）と表し、時刻ｋ－１における個数分布Ｃをｐｋ－１（ｎ）と表すものとする。時刻ｋにおける個数分布の予測値は、ｐｋ｜ｋ－１（ｎ）と表すものとする。時刻ｋにおけるＰＨＤ分布Ｄをｖｋ（ｘ）と表し、時刻ｋ－１におけるＰＨＤ分布Ｄをｖｋ－１（ｘ）と表すものとする。時刻ｋにおけるＰＨＤ分布の予測値はｖｋ｜ｋ－１（ｘ）と表すものとする。 Further, the number distribution C at time k is represented by p k (n), and the number distribution C at time k-1 is represented by p k-1 (n). The predicted value of the number distribution at time k shall be expressed as pk | k-1 (n). The PHD distribution D at time k is represented by v k (x), and the PHD distribution D at time k-1 is represented by v k-1 (x). The predicted value of the PHD distribution at time k shall be expressed as v k | k-1 (x).
　任意の関数ｆおよびｇの積分計算を、＜ｆ，ｇ＞と表し、ｆおよびｇが０以上の整数ｎに対するｆ（ｎ）およびｇ（ｎ）である場合、＜ｆ，ｇ＞は、下記式（１）で表される。

Figure JPOXMLDOC01-appb-I000001
The integral calculation of arbitrary functions f and g is expressed as <f, g>, and when f and g are f (n) and g (n) for an integer n of 0 or more, <f, g> is as follows. It is represented by the equation (1).

Figure JPOXMLDOC01-appb-I000001
　ｆおよびｇが、実数ベクトルである状態ベクトルｘに対する関数ｆ（ｘ）およびｇ（ｘ）である場合、＜ｆ，ｇ＞は、下記式（２）で表される。なお、実数ベクトルに関する積分は、実数ベクトルのベクトル空間全体における積分であるものとする。

Figure JPOXMLDOC01-appb-I000002
When f and g are functions f (x) and g (x) with respect to the state vector x which is a real number vector, <f, g> is represented by the following equation (2). The integral of the real number vector is assumed to be the integral of the real number vector in the entire vector space.

Figure JPOXMLDOC01-appb-I000002
　次に、目標推定装置１の動作について説明する。
　図２は、実施の形態１に係る方法を示すフローチャートであって、目標推定装置１が目標の個数および状態を推定するまでの一連の処理を示している。
　予測処理部１１が、目標の個数分布およびＰＨＤ分布の予測処理を実行する（ステップＳＴ１）。例えば、予測処理部１１が、更新処理部１３によって算出された時刻ｋ－１における個数分布ｐｋ－１（ｎ）およびＰＨＤ分布ｖｋ－１（ｎ）を用いて、下記式（３）および下記式（４）に従う運動モデルに基づき、時刻ｋにおける個数分布の予測値ｐｋ｜ｋ－１（ｎ）と時刻ｋにおけるＰＨＤ分布の予測値ｖｋ｜ｋ－１（ｎ）を算出する。なお、下記式（３）におけるΠｋ｜ｋ－１［ｖ，ｐ］（ｊ）は、下記式（５）で表される。

Figure JPOXMLDOC01-appb-I000003
Next, the operation of the target estimation device 1 will be described.
FIG. 2 is a flowchart showing the method according to the first embodiment, and shows a series of processes until the target estimation device 1 estimates the number and the state of the targets.
The prediction processing unit 11 executes prediction processing of the target number distribution and the PHD distribution (step ST1). For example, the prediction processing unit 11 uses the number distribution p k-1 (n) and the PHD distribution v k-1 (n) at the time k-1 calculated by the update processing unit 13 to use the following equation (3) and Based on the motion model according to the following equation (4), the predicted value of the number distribution at time k | k-1 (n) and the predicted value of the PHD distribution at time k v k | k-1 (n) are calculated. In addition, Π k | k-1 [v, p] (j) in the following formula (3) is represented by the following formula (5).

Figure JPOXMLDOC01-appb-I000003
　上記式（３）および（４）において、ｐΓ，ｋ（ｎ）は、時刻ｋに出現した目標の個数分布に相当するパラメタであり、γｋ（ｘ）は、時刻ｋに出現した目標のＰＨＤ分布に相当するパラメタであり、目標の個数の増加頻度を表すパラメタである。上記式（４）および（５）において、ｐＳ，ｋ｜ｋ－１（ｘ）は、時刻ｋ－１から時刻ｋまでの間に状態ｘの目標が消失せずに残存する確率を表しており、目標の個数の減少頻度を表すパラメタである。上記式（４）におけるｆｋ｜ｋ－１（ｘ｜ζ）は、時刻ｋ－１から時刻ｋまでの間に目標が状態ζから状態ｘへ遷移する確率を表し、目標の運動モデルを表すパラメタである。 In the above equations (3) and (4), p Γ and k (n) are parameters corresponding to the number distribution of targets appearing at time k, and γ k (x) is the target appearing at time k. It is a parameter corresponding to the PHD distribution, and is a parameter indicating the frequency of increase in the number of targets. In the above equations (4) and (5), pS , k | k-1 (x) represents the probability that the target of the state x remains without disappearing between the time k-1 and the time k. This is a parameter that indicates the frequency of decrease in the number of targets. F k | k-1 (x | ζ) in the above equation (4) represents the probability that the target transitions from the state ζ to the state x between the time k-1 and the time k, and represents the motion model of the target. It is a parameter.
　上記式（５）におけるＣｉ ｊは、ｉ個からｊ個を選ぶ場合の数を表す二項係数とする。上記式（４）および（５）における状態ベクトルに関する積分は、混合ガウスモデルまたは逐次モンテカルロ法を用いて、時刻ｋ－１のＰＨＤ分布ｖｋ－１（ｘ）を近似することで計算可能となる。また、上記式（５）における無限級数は、目標の個数の最大値を定め、この最大値よりも十分に大きいｎに対してｐｋ｜ｋ－１（ｎ）＝０とする近似を用いることで計算可能となる。 C i j in the above equation (5) is a binomial coefficient representing the number of cases where the i number choose j number. The integral for the state vector in the above equations (4) and (5) can be calculated by approximating the PHD distribution v k-1 (x) at time k-1 using the mixed Gaussian model or the sequential Monte Carlo method. .. Further, for the infinite series in the above equation (5), the maximum value of the number of targets is determined, and an approximation of pk | k-1 (n) = 0 is used for n which is sufficiently larger than this maximum value. It becomes possible to calculate with.
　次に、個数分布変形部１２が形状パラメタを設定する（ステップＳＴ２）。形状パラメタは、個数分布を変形するために用いられるパラメタである。形状パラメタは、個数分布の形状の過度な広がりまたは過度な狭まりを防いで、変形後の個数分布の分散を指定された範囲内に収めるものである。 Next, the number distribution deformation unit 12 sets the shape parameter (step ST2). The shape parameter is a parameter used to transform the number distribution. The shape parameter prevents the shape of the number distribution from being excessively widened or narrowed, and keeps the variance of the number distribution after deformation within a specified range.
　また、形状パラメタは、変形の前後で個数分布の平均値が同じであるように設定されなければならない。これは、個数分布の平均値が、時刻ｋにおける目標の平均個数を示しており、過去に取得された観測データＡの情報を含んでいることに起因する。すなわち、変形後の個数分布をｐ’ｋ｜ｋ－１（ｎ）とした場合、下記式（６）に示す関係が成り立つ形状パラメタが設定される。

Figure JPOXMLDOC01-appb-I000004
In addition, the shape parameter must be set so that the average value of the number distribution is the same before and after the deformation. This is because the average value of the number distribution indicates the average number of targets at time k and includes the information of the observation data A acquired in the past. That is, when the number distribution after deformation is p'k | k-1 (n), the shape parameters for which the relationship shown in the following equation (6) holds are set.

Figure JPOXMLDOC01-appb-I000004
　形状パラメタは、変形後の個数分布ｐ’ｋ｜ｋ－１（ｎ）に依存する。例えば、変形後の個数分布ｐ’ｋ｜ｋ－１（ｎ）が下記式（７）に従うものとする。下記式（７）において、ｒ、τおよびλが形状パラメタであり、いずれも実数である。形状パラメタｒ、τおよびλが取り得る値の範囲は、下記式（８）から（１０）までに示す範囲である。

Figure JPOXMLDOC01-appb-I000005
The shape parameter depends on the number distribution p'k | k-1 (n) after deformation. For example, it is assumed that the number distribution p'k | k-1 (n) after deformation follows the following equation (7). In the following equation (7), r, τ and λ are shape parameters, and all of them are real numbers. The range of values that the shape parameters r, τ, and λ can take is the range shown in the following equations (8) to (10).

Figure JPOXMLDOC01-appb-I000005
Figure JPOXMLDOC01-appb-I000006
Figure JPOXMLDOC01-appb-I000006
　上記式（７）で表される個数分布ｐ’ｋ｜ｋ－１（ｎ）の平均値μは、下記式（１２）で表される。また、上記式（７）で表される個数分布ｐ’ｋ｜ｋ－１（ｎ）の分散σ２は、下記式（１３）で表される。

Figure JPOXMLDOC01-appb-I000007
The average value μ of the number distribution p'k | k-1 (n) represented by the above formula (7) is represented by the following formula (12). Further, the variance σ 2 of the number distribution p'k | k-1 (n) represented by the above formula (7) is represented by the following formula (13).

Figure JPOXMLDOC01-appb-I000007
　個数分布変形部１２は、上記式（１２）および（１３）に基づいて、下記式（１４）と下記式（１５）を満たす形状パラメタｒ、τおよびλを探索する。下記式（１５）において、σ２ Ｌは、分散の下限値を指定するパラメタであり、σ２ Ｈは、分散の上限値を指定するパラメタである。形状パラメタの探索には、最急降下法といった最適化アルゴリズムを用いることができる。すなわち、変形後の個数分布Ｃの分散σ２に関する指定された範囲は、形状パラメタｒ、τおよびλを用いた下限値σ２ Ｌと上限値σ２ Ｈによって決定される下記式（１５）に示す範囲である。

Figure JPOXMLDOC01-appb-I000008
Based on the above equations (12) and (13), the number distribution deformation unit 12 searches for shape parameters r, τ, and λ that satisfy the following equations (14) and (15). In the following equation (15), σ 2 L is a parameter that specifies the lower limit of the variance, and σ 2 H is a parameter that specifies the upper limit of the variance. An optimization algorithm such as the steepest descent method can be used to search for shape parameters. That is, the specified range for the variance σ 2 of the number distribution C after deformation is determined by the following equation (15) determined by the lower limit value σ 2 L and the upper limit value σ 2 H using the shape parameters r, τ and λ. It is the range shown.

Figure JPOXMLDOC01-appb-I000008
　なお、変形後の個数分布の形状を表す式は上記式（７）に限らず、個数分布変形部１２は、他の式で表された形状を用いることができる。例えば、変形後の個数分布の形状を、下記式（１６）に示す二重ポアソン分布で表してもよい。下記式（１６）において、ｃは規格化定数であり、ｅは自然対数の底である。形状パラメタは、αおよびθである。この場合、変形後の個数分布の形状を決定する形状パラメタとして、変形前の個数分布と同じ平均値を持ち、かつ分散が指定された範囲内に収まる値を求めることができる。ここで、変形後の個数分布Ｃの分散に関する指定された範囲は、形状パラメタαおよびθによって決定される範囲である。

Figure JPOXMLDOC01-appb-I000009
The formula for expressing the shape of the number distribution after deformation is not limited to the above formula (7), and the number distribution deformation unit 12 can use a shape represented by another formula. For example, the shape of the number distribution after deformation may be represented by the double Poisson distribution represented by the following equation (16). In the following equation (16), c is a normalized constant and e is the base of the natural logarithm. The shape parameters are α and θ. In this case, as a shape parameter for determining the shape of the number distribution after deformation, it is possible to obtain a value having the same average value as the number distribution before deformation and having a variance within a specified range. Here, the designated range regarding the variance of the number distribution C after deformation is the range determined by the shape parameters α and θ.

Figure JPOXMLDOC01-appb-I000009
　続いて、個数分布変形部１２は、探索した形状パラメタに基づいて、時刻ｋの予測値である個数分布の形状を変形する（ステップＳＴ３）。例えば、変形後の個数分布が上記式（７）で表される場合、個数分布変形部１２は、形状パラメタｒ、τおよびλを用いて、上記式（７）に従い、時刻ｋの予測値である変形後の個数分布ｐ’ｋ｜ｋ－１（ｎ）を算出する。 Subsequently, the number distribution transforming unit 12 transforms the shape of the number distribution, which is a predicted value at time k, based on the searched shape parameter (step ST3). For example, when the number distribution after transformation is represented by the above equation (7), the number distribution transformation unit 12 uses the shape parameters r, τ and λ, and according to the above equation (7), at the predicted value at time k. The number distribution p'k | k-1 (n) after a certain transformation is calculated.
　更新処理部１３は、時刻ｋにおける目標の個数分布とＰＨＤ分布を算出する更新処理を実行する（ステップＳＴ４）。例えば、更新処理部１３は、個数分布変形部１２によって変形された個数分布の時刻ｋにおける予測値ｐ’ｋ｜ｋ－１（ｎ）と、予測処理部１１によって算出された時刻ｋにおけるＰＨＤ分布の予測値ｖｋ｜ｋ－１（ｘ）とを用いて、下記式（１７）および（１８）に従う観測モデルに基づき、時刻ｋにおける個数分布ｐｋ（ｎ）およびＰＨＤ分布ｖｋ（ｎ）を算出する。なお、下記式（１７）および（１８）におけるΥｕ ｋ［ｖ，Ｚ］（ｎ）は、下記式（１９）で表される。ｕは、０または１である。

Figure JPOXMLDOC01-appb-I000010
The update processing unit 13 executes an update process for calculating the target number distribution and the PHD distribution at time k (step ST4). For example, the update processing unit 13 has a predicted value p'k | k-1 (n) at time k of the number distribution deformed by the number distribution transforming unit 12, and a PHD distribution at time k calculated by the prediction processing unit 11. predicted values v k | using a k-1 (x), the following equation (17) and based on the observation model according to (18), the number distribution p k at time k (n) and PHD distribution v k (n) Is calculated. Incidentally, the following equation (17) and Upsilon u k in (18) [v, Z] (n) is expressed by the following equation (19). u is 0 or 1.

Figure JPOXMLDOC01-appb-I000010
　上記式（１８）および（１９）において、ｐＫ，ｋ（ｘ）は、誤って検出された目標の個数分布を表しており、κｋ（ｘ）は、誤って検出された目標のＰＨＤ分布を表しており、目標の個数に関する観測モデルのパラメタである。ｐＤ，ｋ（ｘ）は、時刻ｋにおいて状態ｘの目標を観測する確率を表しており、同様に目標の個数に関する観測モデルのパラメタである。また、ｇｋ（ｚ｜ｘ）は、時刻ｋにおける状態ｘの目標に対する観測ベクトルｚの尤度を表しており、目標の状態に関する観測モデルのパラメタである。 In the above equations (18) and (19), p K and k (x) represent the number distribution of erroneously detected targets, and κ k (x) is the PHD distribution of erroneously detected targets. Is a parameter of the observation model regarding the number of targets. p D, k (x) represent the probability of observing a target in the state x at time k, and is also a parameter of the observation model regarding the number of targets. Further, g k (z | x) represents the likelihood of the observation vector z with respect to the target of the state x at time k, and is a parameter of the observation model regarding the target state.
　上記式（１８）において、Ｚｋ＼｛ｚ｝は、集合Ｚｋから集合｛ｚ｝を除いた差集合である。また、上記式（１９）において、εｊ（Ａ）は、集合Ａの各要素に関する次数ｊの基本対称式であり、下記式（２０）によって表される。集合Ξｋ（ｖ，Ｚ）は、下記式（２１）で表され、Ｐｎ ｊは、下記式（２２）で表される。

Figure JPOXMLDOC01-appb-I000011
In the above equation (18), Z k \ {z} is a set of differences obtained by removing the set {z} from the set Z k . Further, in the above equation (19), ε j (A) is a basic symmetric equation of order j with respect to each element of the set A, and is represented by the following equation (20). The set Ξ k (v, Z) is represented by the following formula (21), and P n j is represented by the following formula (22).

Figure JPOXMLDOC01-appb-I000011
　上記式（１９）および（２０）における状態ベクトルに関する積分は、混合ガウスモデルまたは逐次モンテカルロ法を用いて、時刻ｋのＰＨＤ分布の予測値ｖｋ｜ｋ－１（ｘ）を近似することで計算可能となる。 The integral for the state vector in the above equations (19) and (20) is calculated by approximating the predicted value v k | k-1 (x) of the PHD distribution at time k using a mixed Gaussian model or the sequential Monte Carlo method. It will be possible.
　推定処理部１４は、時刻ｋにおける目標の個数および状態の推定処理を実行する（ステップＳＴ５）。例えば、推定処理部１４は、更新処理部１３によって算出された時刻ｋにおける目標の個数分布ｐｋ（ｎ）およびＰＨＤ分布ｖｋ（ｘ）を用いて、下記式（２３）に従い、目標の個数の推定値ｎｋを算出する。さらに、推定処理部１４は、目標の個数の推定値ｎｋが１以上であった場合、ＰＨＤ分布ｖｋ（ｘ）のピークのうち、極値が大きい上位ｎｋ個のピークを与える状態ベクトルを求め、この状態ベクトルを目標の状態の推定値とする。例えば、推定処理部１４は、ＰＨＤ分布ｖｋ（ｘ）を混合ガウスモデルまたは逐次モンテカルロ法によって近似し、近似したＰＨＤ分布ｖｋ（ｘ）から、極値が大きい上位ｎｋ個のピークを与える状態ベクトルを求める。

Figure JPOXMLDOC01-appb-I000012
The estimation processing unit 14 executes estimation processing of the number of targets and the state at time k (step ST5). For example, the estimation processing unit 14 uses the number distribution of the target at time k calculated by the update processing unit 13 p k (n) and PHD distribution v k (x), according to the following equation (23), the number of target The estimated value n k of is calculated. Further, the estimation processing unit 14 gives a state vector that gives the upper nk peaks having the largest extreme values among the peaks of the PHD distribution v k (x) when the estimated value nk of the number of targets is 1 or more. Is obtained, and this state vector is used as an estimated value of the target state. For example, the estimation processing unit 14 approximates the PHD distribution v k (x) by a mixed Gaussian model or a sequential Monte Carlo method, and gives the upper nk peaks having a large extremum from the approximated PHD distribution v k (x). Find the state vector.

Figure JPOXMLDOC01-appb-I000012
　なお、これまでの説明では、時刻ｋ－１における目標の個数分布ＣおよびＰＨＤ分布Ｄと、時刻ｋにおける目標の状態の観測データＡに基づいて、時刻ｋにおける目標の個数と状態を推定する処理を示したが、目標の個数および状態を推定するときの時間は必ずしも未来に向かって流れる場合に限らず、過去に向かって流れてもよい。例えば、目標推定装置１は、時刻ｋ＋１における目標の個数分布ＣおよびＰＨＤ分布Ｄと、次の時刻である時刻ｋにおける目標の状態の観測データＡに基づいて、時刻ｋにおける目標の個数および状態を推定してもよい。この場合、予測処理における目標の個数の変動と目標の状態の遷移とを示すパラメタを、時刻ｋ＋１から時刻ｋへの遷移を表すモデルのパラメタとする。 In the description so far, the process of estimating the number of targets and the state at time k based on the target number distribution C and PHD distribution D at time k-1 and the observation data A of the target state at time k. However, the time for estimating the number of targets and the state does not necessarily flow toward the future, but may flow toward the past. For example, the target estimation device 1 determines the number and states of targets at time k based on the target number distribution C and PHD distribution D at time k + 1 and the observation data A of the target state at time k, which is the next time. You may estimate. In this case, the parameter indicating the fluctuation of the number of targets and the transition of the target state in the prediction process is used as the parameter of the model representing the transition from time k + 1 to time k.
　次に、目標推定装置１の機能を実現するハードウェア構成について説明する。
　目標推定装置１における予測処理部１１、個数分布変形部１２、更新処理部１３および推定処理部１４の各機能は、処理回路により実現される。すなわち、目標推定装置１は、図２のステップＳＴ１からステップＳＴ５までの処理を実行する処理回路を備えている。処理回路は、専用のハードウェアであってもよいし、メモリに記憶されたプログラムを実行するＣＰＵ（Ｃｅｎｔｒａｌ　Ｐｒｏｃｅｓｓｉｎｇ　Ｕｎｉｔ）であってもよい。 Next, the hardware configuration that realizes the function of the target estimation device 1 will be described.
Each function of the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14 in the target estimation device 1 is realized by the processing circuit. That is, the target estimation device 1 includes a processing circuit that executes the processes from step ST1 to step ST5 in FIG. The processing circuit may be dedicated hardware or a CPU (Central Processing Unit) that executes a program stored in the memory.
　図３Ａは、目標推定装置１の機能を実現するハードウェア構成を示すブロック図であり、図３Ｂは、目標推定装置１の機能を実現するソフトウェアを実行するハードウェア構成を示すブロック図である。図３Ａおよび図３Ｂにおいて、入力インタフェース１００は、観測データ記憶部２から目標推定装置１へ入力される目標の状態の観測データＡを中継するインタフェースである。出力インタフェース１０１は、目標推定装置１から推定値記憶部３へ出力される目標の個数および状態の推定値Ｂを中継するインタフェースである。 FIG. 3A is a block diagram showing a hardware configuration that realizes the function of the target estimation device 1, and FIG. 3B is a block diagram showing a hardware configuration that executes software that realizes the function of the target estimation device 1. In FIGS. 3A and 3B, the input interface 100 is an interface that relays the observation data A of the target state input from the observation data storage unit 2 to the target estimation device 1. The output interface 101 is an interface that relays the estimated value B of the number of targets and the state output from the target estimation device 1 to the estimated value storage unit 3.
　入力インタフェース１００および出力インタフェース１０１は、例えば、ＤＶＩ（登録商標）（Ｄｉｇｉｔａｌ　Ｖｉｓｕａｌ　Ｉｎｔｅｒｆａｃｅ）、ＨＤＭＩ（登録商標）（Ｈｉｇｈｔ－Ｄｅｆｉｎｉｔｉｏｎ　Ｍｕｌｔｉｍｅｄｉａ　Ｉｎｔｅｒｆａｃｅ）、ＵＳＢ（Ｕｎｉｖｅｒｓａｌ　Ｓｅｒｉａｌ　Ｂｕｓ）、イーサネット（登録商標）、またはＣＡＮ（Ｃｏｎｔｒｏｌｌｅｒ　Ａｒｅａ　Ｎｅｔｗｏｒｋ）バスであってもよい。 The input interface 100 and the output interface 101 are, for example, DVI (registered trademark) (Digital Visual Interface), HDMI (registered trademark) (High-Definition Multimedia Interface), USB (Universal Serial Bus), Ethernet (registered trademark), or Ethernet (registered trademark). It may be a (Control Area Network) bus.
　処理回路が図３Ａに示す専用のハードウェアの処理回路１０２である場合に、処理回路１０２は、例えば、単一回路、複合回路、プログラム化したプロセッサ、並列プログラム化したプロセッサ、ＡＳＩＣ（Ａｐｐｌｉｃａｔｉｏｎ　Ｓｐｅｃｉｆｉｃ　Ｉｎｔｅｇｒａｔｅｄ　Ｃｉｒｃｕｉｔ）、ＦＰＧＡ（Ｆｉｅｌｄ－Ｐｒｏｇｒａｍｍａｂｌｅ　Ｇａｔｅ　Ａｒｒａｙ）またはこれらを組み合わせたものが該当する。目標推定装置１における予測処理部１１、個数分布変形部１２、更新処理部１３および推定処理部１４の機能を、別々の処理回路で実現してもよいし、これらの機能をまとめて１つの処理回路で実現してもよい。 When the processing circuit is the processing circuit 102 of the dedicated hardware shown in FIG. 3A, the processing circuit 102 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, or an ASIC (Application Specific Integrated Circuitd). Circuit), FPGA (Field-Programmable Gate Array), or a combination thereof is applicable. The functions of the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14 in the target estimation device 1 may be realized by separate processing circuits, or these functions may be combined into one process. It may be realized by a circuit.
　処理回路が図３Ｂに示すプロセッサ１０３である場合に、目標推定装置１における予測処理部１１、個数分布変形部１２、更新処理部１３および推定処理部１４の機能は、ソフトウェア、ファームウェアまたはソフトウェアとファームウェアとの組み合わせによって実現される。なお、ソフトウェアまたはファームウェアは、プログラムとして記述されてメモリ１０４に記憶される。 When the processing circuit is the processor 103 shown in FIG. 3B, the functions of the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14 in the target estimation device 1 are software, firmware, or software and firmware. It is realized by the combination with. The software or firmware is described as a program and stored in the memory 104.
　プロセッサ１０３は、メモリ１０４に記憶されたプログラムを読み出して実行することによって、目標推定装置１における予測処理部１１、個数分布変形部１２、更新処理部１３および推定処理部１４の機能を実現する。例えば、目標推定装置１は、プロセッサ１０３によって実行されるときに、図２に示すフローチャートにおける、ステップＳＴ１からステップＳＴ５の処理が結果的に実行されるプログラムを記憶するためのメモリ１０４を備える。これらのプログラムは、予測処理部１１、個数分布変形部１２、更新処理部１３および推定処理部１４の手順または方法をコンピュータに実行させる。メモリ１０４は、コンピュータを、予測処理部１１、個数分布変形部１２、更新処理部１３および推定処理部１４として機能させるためのプログラムが記憶されたコンピュータ可読記憶媒体であってもよい。 The processor 103 realizes the functions of the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14 in the target estimation device 1 by reading and executing the program stored in the memory 104. For example, the target estimation device 1 includes a memory 104 for storing a program in which the processes of steps ST1 to ST5 in the flowchart shown in FIG. 2 are executed as a result when executed by the processor 103. These programs cause a computer to execute the procedures or methods of the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14. The memory 104 may be a computer-readable storage medium in which a program for causing the computer to function as the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14 is stored.
　メモリ１０４は、例えば、ＲＡＭ（Ｒａｎｄｏｍ　Ａｃｃｅｓｓ　Ｍｅｍｏｒｙ）、ＲＯＭ（Ｒｅａｄ　Ｏｎｌｙ　Ｍｅｍｏｒｙ）、フラッシュメモリ、ＥＰＲＯＭ（Ｅｒａｓａｂｌｅ　Ｐｒｏｇｒａｍｍａｂｌｅ　Ｒｅａｄ　Ｏｎｌｙ　Ｍｅｍｏｒｙ）、ＥＥＰＲＯＭ（Ｅｌｅｃｔｒｉｃａｌｌｙ－ＥＰＲＯＭ）などの不揮発性または揮発性の半導体メモリ、磁気ディスク、フレキシブルディスク、光ディスク、コンパクトディスク、ミニディスク、ＤＶＤなどが該当する。 The memory 104 is, for example, a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory), an EEPROM (Electrically-volatile) semiconductor, an EPROM (Electrically-volatile), or the like. This includes disks, flexible disks, optical disks, compact disks, mini disks, DVDs, and the like.
　目標推定装置１における予測処理部１１、個数分布変形部１２、更新処理部１３および推定処理部１４の機能の一部を専用のハードウェアで実現し、一部をソフトウェアまたはファームウェアで実現してもよい。例えば、予測処理部１１、更新処理部１３および推定処理部１４は、専用のハードウェアである処理回路１０２によって機能を実現し、個数分布変形部１２は、プロセッサ１０３がメモリ１０４に記憶されたプログラムを読み出して実行することによって、機能を実現する。このように、処理回路はハードウェア、ソフトウェア、ファームウェアまたはこれらの組み合わせにより上記機能を実現することができる。 Even if some of the functions of the prediction processing unit 11, the number distribution deformation unit 12, the update processing unit 13, and the estimation processing unit 14 in the target estimation device 1 are realized by dedicated hardware, and some of them are realized by software or firmware. Good. For example, the prediction processing unit 11, the update processing unit 13, and the estimation processing unit 14 realize the functions by the processing circuit 102, which is dedicated hardware, and the number distribution transformation unit 12 is a program in which the processor 103 is stored in the memory 104. The function is realized by reading and executing. In this way, the processing circuit can realize the above functions by hardware, software, firmware or a combination thereof.
　以上のように、実施の形態１に係る目標推定装置１は、予測処理部１１、個数分布変形部１２、更新処理部１３および推定処理部１４を備える。推定処理部１４が、時刻ｋにおける目標の個数分布ＣおよびＰＨＤ分布Ｄに基づいて目標の個数および状態を推定するにあたり、個数分布変形部１２が、分布の分散が指定した範囲内に収まるように、時刻ｋにおける個数分布の予測値を変形する。これにより、目標の観測データの誤検出および欠落が頻発する場合であっても、逐次算出される個数分布が過度に広がることを防止できる。結果として、目標の個数分布の分散が大きいことに起因して目標の個数の推定値が不安定になる事態が発生する頻度を減らすことが可能である。また、観測データＡのいずれかの精度が過大に評価されていた場合、すなわち観測データＡに設定された観測誤差、誤検出頻度または観測データＡが欠落する頻度が実際の観測誤差、実際の誤差検出頻度または実際の観測データＡが欠落する頻度よりも小さく設定された場合であっても、逐次算出される目標の個数分布が過度に狭まる事態を防止できる。結果として、目標の個数分布の分散が小さいことに起因する、目標の個数の変動に対して目標の個数の推定値が遅延する事態の頻度を減らすことができる。このように、目標推定装置１は、目標の状態量の推定値と、時間経過に対して安定した目標の個数の推定値とを算出することができる。 As described above, the target estimation device 1 according to the first embodiment includes a prediction processing unit 11, a number distribution deformation unit 12, an update processing unit 13, and an estimation processing unit 14. When the estimation processing unit 14 estimates the number and state of the target based on the target number distribution C and the PHD distribution D at time k, the number distribution deformation unit 12 so that the variance of the distribution falls within the specified range. , Transform the predicted value of the number distribution at time k. As a result, even when erroneous detection and omission of the target observation data occur frequently, it is possible to prevent the number distribution calculated sequentially from expanding excessively. As a result, it is possible to reduce the frequency of instability of the estimated number of targets due to the large variance of the number distribution of the targets. In addition, when the accuracy of any of the observation data A is overestimated, that is, the observation error set in the observation data A, the false detection frequency, or the frequency at which the observation data A is missing is the actual observation error or the actual error. Even when the detection frequency or the frequency at which the actual observation data A is missing is set to be smaller, it is possible to prevent a situation in which the number distribution of the targets calculated sequentially is excessively narrowed. As a result, it is possible to reduce the frequency of delays in the estimated number of targets with respect to fluctuations in the number of targets due to the small variance of the number distribution of the targets. In this way, the target estimation device 1 can calculate an estimated value of the target state quantity and an estimated value of the number of targets that are stable over time.
実施の形態２．
　実施の形態１では、目標の個数分布が望ましい広がりを持つように、予測値の個数分布の形状を変形する処理を行った。これに対して、実施の形態２では、個数分布の予測処理に用いられる個数分布の形状を変形する。実施の形態２においても、目標の個数分布は、予測処理と更新処理で再帰的に用いられるため、実施の形態１と同様の効果が得られる。 Embodiment 2.
In the first embodiment, a process of deforming the shape of the number distribution of the predicted values is performed so that the target number distribution has a desired spread. On the other hand, in the second embodiment, the shape of the number distribution used in the number distribution prediction process is modified. Also in the second embodiment, since the target number distribution is recursively used in the prediction process and the update process, the same effect as that of the first embodiment can be obtained.
　また、実施の形態１における予測処理部１１によって予測処理に用いられる目標の個数の変動モデルを表すパラメタは、例えば、上記式（３）および（４）における目標の増加頻度を表すパラメタｐΓ，ｋ（ｎ）およびγｋ（ｘ）である。これらのパラメタの値は、個数分布の平均値のみに反映され、個数分布の形状には反映されない。 Further, the parameters representing the fluctuation model of the number of targets used in the prediction processing by the prediction processing unit 11 in the first embodiment are, for example, the parameters p Γ, which represent the increase frequency of the targets in the above equations (3) and (4) . k (n) and γ k (x). The values of these parameters are reflected only in the mean value of the number distribution, not in the shape of the number distribution.
　これらのパラメタの値の変更が目標の個数および状態の推定結果に与える影響は、既存のＣＰＨＤフィルタに比べて小さい。この特性は、目標の個数の変動モデルが実際の目標の個数の変動を極めて正確に表していても、既存のＣＰＨＤフィルタに比べて目標の個数および状態の推定精度が劣化する頻度を増加させる要因となる。 The effect of changing the values of these parameters on the target number and state estimation results is smaller than that of existing CPHD filters. This characteristic is a factor that increases the frequency of deterioration in the estimation accuracy of the number of targets and the state compared to the existing CPHD filter, even if the variation model of the number of targets represents the variation of the actual number of targets extremely accurately. It becomes.
　そこで、実施の形態２では、予測処理部１１Ａが、個数分布変形部１２Ａによって変形された個数分布を用いて個数分布の予測値を算出し、更新処理部１３に入力する。この構成により、実施の形態１と同様の効果が得られることに加え、目標の個数の変動モデルを表すパラメタを、目標の個数および状態の推定結果により強く反映させることができる。 Therefore, in the second embodiment, the prediction processing unit 11A calculates the predicted value of the number distribution using the number distribution deformed by the number distribution deformation unit 12A, and inputs it to the update processing unit 13. With this configuration, in addition to obtaining the same effect as in the first embodiment, the parameters representing the variation model of the number of targets can be more strongly reflected in the estimation result of the number of targets and the state.
　図４は、実施の形態２に係る目標推定装置１Ａの構成を示すブロック図である。図４において、図１と同一の構成要素には同一の符号を付して説明を省略する。目標推定装置１Ａは、実施の形態１と同様に、観測データ記憶部２に記憶された目標の状態の観測データＡを用いて、目標の個数および状態の推定値Ｂを算出して推定値記憶部３に記憶する。目標推定装置１Ａは、図４に示すように、予測処理部１１Ａ、個数分布変形部１２Ａ、更新処理部１３および推定処理部１４を備える。 FIG. 4 is a block diagram showing the configuration of the target estimation device 1A according to the second embodiment. In FIG. 4, the same components as those in FIG. 1 are designated by the same reference numerals, and the description thereof will be omitted. Similar to the first embodiment, the target estimation device 1A calculates the number of targets and the estimated value B of the state by using the observation data A of the target state stored in the observation data storage unit 2, and stores the estimated value. Store in part 3. As shown in FIG. 4, the target estimation device 1A includes a prediction processing unit 11A, a number distribution deformation unit 12A, an update processing unit 13, and an estimation processing unit 14.
　予測処理部１１Ａは、個数分布変形部１２Ａによって変形された時刻ｋ－１における目標の個数分布Ｃｂと、更新処理部１３によって算出された時刻ｋ－１における目標のＰＨＤ分布Ｄに基づいて、時刻ｋにおける個数分布の予測値ＣａｂとＰＨＤ分布の予測値Ｄａを算出する。予測処理部１１Ａによって算出された時刻ｋにおける個数分布の予測値ＣａｂおよびＰＨＤ分布の予測値Ｄａは、更新処理部１３に出力される。更新処理部１３は、時刻ｋにおける個数分布の予測値Ｃａｂ、時刻ｋにおけるＰＨＤ分布の予測値Ｄａおよび時刻ｋにおける目標の状態の観測データＡに基づいて、時刻ｋにおける個数分布ＣおよびＰＨＤ分布Ｄを算出する。 The prediction processing unit 11A is based on the target number distribution Cb at the time k-1 deformed by the number distribution transforming unit 12A and the target PHD distribution D at the time k-1 calculated by the update processing unit 13. The predicted value Cab of the number distribution in k and the predicted value Da of the PHD distribution are calculated. The predicted value Cab of the number distribution and the predicted value Da of the PHD distribution at time k calculated by the prediction processing unit 11A are output to the update processing unit 13. The update processing unit 13 determines the number distribution C and the PHD distribution D at time k based on the predicted value Cab of the number distribution at time k, the predicted value Da of the PHD distribution at time k, and the observation data A of the target state at time k. Is calculated.
　個数分布変形部１２Ａは、更新処理部１３によって算出された時刻ｋ－１における個数分布Ｃを、当該個数分布Ｃの分散が指定した範囲内に収まるように変形する分布変形部である。個数分布変形部１２Ａによって分布が変形された時刻ｋ－１における個数分布Ｃｂは、予測処理部１１Ａに出力される。 The number distribution deformation unit 12A is a distribution deformation unit that transforms the number distribution C at time k-1 calculated by the update processing unit 13 so that the variance of the number distribution C falls within a designated range. The number distribution Cb at time k-1 in which the distribution is deformed by the number distribution transforming unit 12A is output to the prediction processing unit 11A.
　次に、目標推定装置１Ａの動作について説明する。
　図５は、実施の形態２に係る方法を示すフローチャートであり、目標推定装置１Ａが目標の個数および状態を推定するまでの一連の処理を示している。
　個数分布変形部１２Ａが形状パラメタを設定する（ステップＳＴ１ａ）。形状パラメタは、実施の形態１と同様に、個数分布を変形するために用いられるパラメタである。形状パラメタは、個数分布の形状の過度な広がりまたは過度な狭まりを防いで、変形後の個数分布の分散を指定された範囲内に収めるものである。また、形状パラメタは、変形の前後で個数分布の平均値が同じであるように設定されなければならない。これは、個数分布の平均値が、時刻ｋにおける目標の平均個数を示しており、過去に取得された観測データＡの情報を含んでいることに起因する。 Next, the operation of the target estimation device 1A will be described.
FIG. 5 is a flowchart showing the method according to the second embodiment, and shows a series of processes until the target estimation device 1A estimates the number and the state of the targets.
The number distribution deformation unit 12A sets the shape parameter (step ST1a). The shape parameter is a parameter used to transform the number distribution, as in the first embodiment. The shape parameter prevents the shape of the number distribution from being excessively widened or narrowed, and keeps the variance of the number distribution after deformation within a specified range. In addition, the shape parameter must be set so that the average value of the number distribution is the same before and after the deformation. This is because the average value of the number distribution indicates the average number of targets at time k and includes the information of the observation data A acquired in the past.
　個数分布変形部１２Ａは、上記式（１２）および（１３）に基づいて、上記式（１４）と上記式（１５）を満たす形状パラメタｒ、τおよびλを探索する。形状パラメタの探索には、最急降下法といった最適化アルゴリズムを用いることができる。すなわち、変形後の個数分布Ｃの分散σ２に関する指定された範囲は、形状パラメタｒ、τおよびλを用いた下限値σ２ Ｌと上限値σ２ Ｈによって決定される上記式（１５）に示す範囲である。 Based on the above equations (12) and (13), the number distribution deformation unit 12A searches for shape parameters r, τ, and λ that satisfy the above equations (14) and (15). An optimization algorithm such as the steepest descent method can be used to search for shape parameters. That is, the specified range for the variance σ 2 of the number distribution C after deformation is determined by the above equation (15) determined by the lower limit value σ 2 L and the upper limit value σ 2 H using the shape parameters r, τ and λ. It is the range shown.
　変形後の個数分布の形状を表す式は上記式（７）に限らず、個数分布変形部１２Ａは、他の式で表された形状を用いることができる。例えば、変形後の個数分布の形状を、上記式（１６）に示す二重ポアソン分布で表してもよい。上記式（１６）において、ｃは規格化定数であり、ｅは自然対数の底である。形状パラメタはαおよびθである。この場合、変形後の個数分布の形状を決定する形状パラメタとして、変形前の個数分布と同じ平均値を持ち、かつ分散が指定された範囲内に収まる値を求めることができる。ここで、変形後の個数分布Ｃの分散に関する指定された範囲は、形状パラメタαおよびθによって決定される範囲である。 The formula for expressing the shape of the number distribution after deformation is not limited to the above formula (7), and the number distribution deformation unit 12A can use a shape represented by another formula. For example, the shape of the number distribution after deformation may be represented by the double Poisson distribution represented by the above equation (16). In the above equation (16), c is a normalized constant and e is the base of the natural logarithm. The shape parameters are α and θ. In this case, as a shape parameter for determining the shape of the number distribution after deformation, it is possible to obtain a value having the same average value as the number distribution before deformation and having a variance within a specified range. Here, the designated range regarding the variance of the number distribution C after deformation is the range determined by the shape parameters α and θ.
　個数分布変形部１２Ａは、探索した形状パラメタに基づいて、時刻ｋ－１における個数分布の形状を変形する（ステップＳＴ２ａ）。例えば、変形後の個数分布が上記式（７）で表される場合、個数分布変形部１２Ａは、形状パラメタｒ、τおよびλを用いて、上記式（７）に従い、時刻ｋ－１における変形後の個数分布Ｃｂを算出する。 The number distribution deformation unit 12A deforms the shape of the number distribution at time k-1 based on the searched shape parameter (step ST2a). For example, when the number distribution after deformation is represented by the above formula (7), the number distribution deformation unit 12A uses the shape parameters r, τ and λ and is deformed at time k-1 according to the above formula (7). The latter number distribution Cb is calculated.
　続いて、予測処理部１１Ａが、目標の個数分布およびＰＨＤ分布の予測処理を実行する（ステップＳＴ３ａ）。例えば、予測処理部１１Ａが、個数分布変形部１２Ａによって変形された時刻ｋ－１における個数分布と、更新処理部１３によって算出された時刻ｋ－１におけるＰＨＤ分布とを用いて、上記式（３）および上記式（４）に従った運動モデルに基づき、時刻ｋにおける個数分布の予測値およびＰＨＤ分布の予測値を算出する。 Subsequently, the prediction processing unit 11A executes the prediction processing of the target number distribution and the PHD distribution (step ST3a). For example, the prediction processing unit 11A uses the number distribution at time k-1 deformed by the number distribution transforming unit 12A and the PHD distribution at time k-1 calculated by the update processing unit 13 to use the above equation (3). ) And the predicted value of the number distribution and the predicted value of the PHD distribution at time k are calculated based on the motion model according to the above equation (4).
　更新処理部１３は、時刻ｋにおける目標の個数分布とＰＨＤ分布を算出する更新処理を実行する（ステップＳＴ４ａ）。例えば、更新処理部１３は、予測処理部１１Ａによって算出された時刻ｋにおける個数分布の予測値およびＰＨＤ分布の予測値を用いて、上記式（１７）および（１８）に従う観測モデルに基づき、時刻ｋにおける目標の個数分布およびＰＨＤ分布を算出する。 The update processing unit 13 executes an update process for calculating the target number distribution and PHD distribution at time k (step ST4a). For example, the update processing unit 13 uses the predicted value of the number distribution and the predicted value of the PHD distribution at time k calculated by the prediction processing unit 11A, and the time is based on the observation model according to the above equations (17) and (18). Calculate the target number distribution and PHD distribution in k.
　推定処理部１４は、時刻ｋにおける目標の個数および状態の推定処理を実行する（ステップＳＴ５ａ）。例えば、推定処理部１４は、更新処理部１３によって算出された時刻ｋにおける目標の個数分布およびＰＨＤ分布を用いて、上記式（２３）に従い、目標の個数の推定値ｎｋを算出する。推定処理部１４は、目標の個数の推定値ｎｋが１以上であった場合、ＰＨＤ分布のピークのうち、極値が大きい上位ｎｋ個のピークを与える状態ベクトルを求め、この状態ベクトルを目標の状態の推定値とする。例えば、推定処理部１４は、ＰＨＤ分布を混合ガウスモデルまたは逐次モンテカルロ法によって近似し、近似したＰＨＤ分布から、極値が大きい上位ｎｋ個のピークを与える状態ベクトルを求める。 The estimation processing unit 14 executes estimation processing of the number of targets and the state at time k (step ST5a). For example, the estimation processing unit 14 calculates an estimated value n k of the number of targets according to the above equation (23) using the target number distribution and the PHD distribution at time k calculated by the update processing unit 13. When the estimated value n k of the number of targets is 1 or more, the estimation processing unit 14 obtains a state vector that gives the upper nk peaks having a large extreme value among the peaks of the PHD distribution, and obtains this state vector. It is an estimate of the target state. For example, the estimation processing unit 14 approximates the PHD distribution by a mixed Gaussian model or a sequential Monte Carlo method, and obtains a state vector that gives the upper nk peaks having large extreme values from the approximated PHD distribution.
　これまでの説明では、時刻ｋ－１における目標の個数分布ＣおよびＰＨＤ分布Ｄと、時刻ｋにおける目標の状態の観測データＡに基づいて、時刻ｋにおける目標の個数と状態を推定する処理を示したが、目標の個数および状態を推定するときの時間は必ずしも未来に向かって流れる場合に限らず、過去に向かって流れる場合もある。例えば、目標推定装置１Ａは、時刻ｋ＋１における目標の個数分布ＣおよびＰＨＤ分布Ｄと、時刻ｋにおける目標の状態の観測データＡに基づいて、時刻ｋにおける目標の個数および状態を推定してもよい。なお、この場合、予測処理における目標の個数の変動および目標の状態の遷移を示すパラメタを、時刻ｋ＋１から時刻ｋへの遷移を表すモデルに関するパラメタとする。 In the explanation so far, the process of estimating the number of targets and the state at time k is shown based on the target number distribution C and PHD distribution D at time k-1 and the observation data A of the target state at time k. However, the time for estimating the number and state of targets does not necessarily flow toward the future, but may flow toward the past. For example, the target estimation device 1A may estimate the number and state of targets at time k based on the number distribution C and PHD distribution D of targets at time k + 1 and the observation data A of the state of the target at time k. .. In this case, the parameter indicating the change in the number of targets and the transition of the target state in the prediction process is used as the parameter related to the model indicating the transition from time k + 1 to time k.
　なお、目標推定装置１Ａにおける、予測処理部１１Ａ、個数分布変形部１２Ａ、更新処理部１３および推定処理部１４の機能は、処理回路によって実現される。すなわち、目標推定装置１Ａは、図４に示したステップＳＴ１ａからステップＳＴ５ａの処理を実行するための処理回路を備える。処理回路は、図３Ａに示した専用のハードウェアの処理回路１０２であってもよいし、図３Ｂに示したメモリ１０４に記憶されたプログラムを実行するプロセッサ１０３であってもよい。 The functions of the prediction processing unit 11A, the number distribution deformation unit 12A, the update processing unit 13, and the estimation processing unit 14 in the target estimation device 1A are realized by the processing circuit. That is, the target estimation device 1A includes a processing circuit for executing the processing of steps ST1a to ST5a shown in FIG. The processing circuit may be the processing circuit 102 of the dedicated hardware shown in FIG. 3A, or the processor 103 that executes the program stored in the memory 104 shown in FIG. 3B.
　以上のように、実施の形態２に係る目標推定装置１Ａは、予測処理部１１Ａ、個数分布変形部１２Ａ、更新処理部１３および推定処理部１４を備える。推定処理部１４が、時刻ｋにおける目標の個数分布ＣおよびＰＨＤ分布Ｄに基づいて目標の個数および状態を推定するにあたり、個数分布変形部１２Ａが、分布の分散が指定した範囲内に収まるように時刻ｋ－１における個数分布Ｃを変形する。予測処理部１１Ａは、個数分布変形部１２Ａによって変形された時刻ｋ－１における目標の個数分布Ｃｂと、更新処理部１３によって算出された時刻ｋ－１における目標のＰＨＤ分布Ｄとに基づいて、時刻ｋにおける個数分布の予測値ＣａｂおよびＰＨＤ分布の予測値Ｄａを算出する。これにより、実施の形態１と同様に、目標の観測データの誤検出および欠落が頻発する場合であっても、逐次算出される個数分布が過度に広がることを防止できる。結果として、目標の個数分布の分散が大きいことに起因して目標の個数の推定値が不安定になる事態が発生する頻度を減らすことができる。また、個数分布の変形が予測処理の前段で実行されるので、目標の個数の変動モデルを表すパラメタを、目標の個数および状態の推定結果により強く反映させることができる。これにより、目標の観測データの誤検出および欠落が頻発する場合であり、かつ目標の個数の変動モデルが、実際の目標の個数の変動を極めて正確に表している場合において、既存のＣＰＨＤフィルタに比べて目標の個数および状態の推定精度が劣化する頻度を減らすことができる。 As described above, the target estimation device 1A according to the second embodiment includes a prediction processing unit 11A, a number distribution deformation unit 12A, an update processing unit 13, and an estimation processing unit 14. When the estimation processing unit 14 estimates the number and state of the target based on the target number distribution C and the PHD distribution D at time k, the number distribution deformation unit 12A is set so that the distribution variance is within the specified range. The number distribution C at time k-1 is transformed. The prediction processing unit 11A is based on the target number distribution Cb at time k-1 deformed by the number distribution transforming unit 12A and the target PHD distribution D at time k-1 calculated by the update processing unit 13. The predicted value Cab of the number distribution at time k and the predicted value Da of the PHD distribution are calculated. As a result, as in the first embodiment, even when erroneous detection and omission of the target observation data occur frequently, it is possible to prevent the number distribution calculated sequentially from expanding excessively. As a result, it is possible to reduce the frequency with which the estimated value of the number of targets becomes unstable due to the large variance of the number distribution of the targets. Further, since the deformation of the number distribution is executed in the first stage of the prediction process, the parameters representing the fluctuation model of the number of targets can be more strongly reflected in the estimation result of the number of targets and the state. As a result, in cases where false detections and omissions of target observation data occur frequently, and when the fluctuation model of the number of targets represents the fluctuation of the actual number of targets extremely accurately, the existing CPHD filter can be used. In comparison, the frequency of deterioration in the number of targets and the estimation accuracy of the state can be reduced.
　なお、本発明は上記実施の形態に限定されるものではなく、本発明の範囲内において、実施の形態のそれぞれの自由な組み合わせまたは実施の形態のそれぞれの任意の構成要素の変形もしくは実施の形態のそれぞれにおいて任意の構成要素の省略が可能である。 The present invention is not limited to the above-described embodiment, and within the scope of the present invention, any combination of the embodiments or any component of the embodiment may be modified or the embodiment. Any component can be omitted in each of the above.
　本発明に係る装置は、例えば、車両、船舶、航空機、ロボット、人または自転車を目標として、目標の個数および状態を推定することができる。 The device according to the present invention can estimate the number and state of targets, for example, targeting vehicles, ships, aircraft, robots, people or bicycles.
　１，１Ａ　目標推定装置、２　観測データ記憶部、３　推定値記憶部、１１，１１Ａ　予測処理部、１２，１２Ａ　個数分布変形部、１３　更新処理部、１４　推定処理部、１００　入力インタフェース、１０１　出力インタフェース、１０２　処理回路、１０３　プロセッサ、１０４　メモリ。 1,1A target estimation device, 2 observation data storage unit, 3 estimation value storage unit, 11,11A prediction processing unit, 12,12A number distribution deformation unit, 13 update processing unit, 14 estimation processing unit, 100 input interface, 101 output Interface, 102 processing circuit, 103 processor, 104 memory.

Claims (6)

  1. 　観測範囲内における目標の個数を表す確率分布である第１の分布と、前記目標の状態を表す状態空間内における前記目標の平均個数の密度を表す分布である第２の分布に基づいて、前記目標の個数および状態を推定する装置であって、
    　前記第１の分布の予測値および前記第２の分布の予測値を算出する予測処理部と、
    　分布の分散が指定した範囲内に収まるように前記第１の分布の予測値または前記第１の分布を変形する分布変形部と、
    　前記第１の分布の予測値、前記第２の分布の予測値および前記目標の運動諸元の観測データに基づいて、前記第１の分布および前記第２の分布を算出する更新処理部と、
    　算出された前記第１の分布および前記第２の分布に基づいて前記目標の個数および状態を逐次的に推定する推定処理部と、
    　を備えたことを特徴とする目標推定装置。     Based on the first distribution, which is a probability distribution representing the number of targets in the observation range, and the second distribution, which is a distribution representing the density of the average number of targets in the state space representing the state of the target. A device that estimates the number and status of targets,
    A prediction processing unit that calculates the predicted value of the first distribution and the predicted value of the second distribution, and
    A distribution deformation part that deforms the predicted value of the first distribution or the first distribution so that the variance of the distribution falls within the specified range.
    An update processing unit that calculates the first distribution and the second distribution based on the predicted value of the first distribution, the predicted value of the second distribution, and the observation data of the motion specifications of the target.
    An estimation processing unit that sequentially estimates the number and state of the target based on the calculated first distribution and the second distribution.
    A target estimation device characterized by being equipped with.
  2. 　前記分布変形部は、前記第１の分布の予測値を変形して前記更新処理部に出力すること
    　を特徴とする請求項１記載の目標推定装置。     The target estimation device according to claim 1, wherein the distribution deformation unit transforms the predicted value of the first distribution and outputs the predicted value to the update processing unit.
  3. 　前記分布変形部は、前記予測処理部による分布の予測値の算出に用いられる前記第１の分布を変形して当該予測処理部に出力すること
    　を特徴とする請求項１記載の目標推定装置。     The target estimation device according to claim 1, wherein the distribution deformation unit deforms the first distribution used for calculating a predicted value of the distribution by the prediction processing unit and outputs the first distribution to the prediction processing unit.
  4. 　前記分布変形部は、変形前後で分布の平均値が同じになるように前記第１の分布を変形すること
    　を特徴とする請求項１から請求項３のいずれか１項記載の目標推定装置。     The target estimation device according to any one of claims 1 to 3, wherein the distribution deformation unit deforms the first distribution so that the average value of the distribution becomes the same before and after the deformation.
  5. 　観測範囲内における目標の個数を表す確率分布である第１の分布と、前記目標の状態を表す状態空間内における前記目標の平均個数の密度を表す分布である第２の分布に基づいて、前記目標の個数および状態を推定する方法であって、
    　予測処理部が、観測時刻の前または後の時刻における前記第１の分布および前記第２の分布を用いて、観測時刻における前記第１の分布の予測値および前記第２の分布の予測値を算出するステップと、
    　分布変形部が、分布の分散が指定した範囲内に収まるように前記第１の分布の予測値を変形するステップと、
    　更新処理部が、観測時刻における変形された前記第１の分布の予測値、観測時刻における前記第２の分布の予測値および観測時刻における前記目標の運動諸元の観測データに基づいて、観測時刻における前記第１の分布および前記第２の分布を算出するステップと、
    　推定処理部が、算出された前記第１の分布および前記第２の分布に基づいて前記目標の個数および状態を観測時刻ごとに推定するステップと、
    　を備えたことを特徴とする目標推定方法。     Based on the first distribution, which is a probability distribution representing the number of targets in the observation range, and the second distribution, which is a distribution representing the density of the average number of targets in the state space representing the state of the target. A method of estimating the number and state of targets,
    The prediction processing unit uses the first distribution and the second distribution at the time before or after the observation time to obtain the predicted value of the first distribution and the predicted value of the second distribution at the observation time. Steps to calculate and
    A step in which the distribution deformation part transforms the predicted value of the first distribution so that the variance of the distribution falls within the specified range.
    The update processing unit performs the observation time based on the predicted value of the first distribution deformed at the observation time, the predicted value of the second distribution at the observation time, and the observation data of the motion specifications of the target at the observation time. And the step of calculating the first distribution and the second distribution in
    A step in which the estimation processing unit estimates the number and state of the targets for each observation time based on the calculated first distribution and the second distribution.
    A target estimation method characterized by being equipped with.
  6. 　観測範囲内における目標の個数を表す確率分布である第１の分布と、前記目標の状態を表す状態空間内における前記目標の平均個数の密度を表す分布である第２の分布に基づいて、前記目標の個数および状態を推定する方法であって、
    　分布変形部が、分布の分散が指定した範囲内に収まるように前記第１の分布を変形するステップと、
    　予測処理部が、変形された前記第１の分布と、前記第２の分布とを用いて、観測時刻における前記第１の分布の予測値および前記第２の分布の予測値を算出するステップと、
    　更新処理部が、観測時刻における前記第１の分布の予測値、観測時刻における前記第２の分布の予測値および観測時刻における前記目標の運動諸元の観測データに基づいて、観測時刻における前記第１の分布および前記第２の分布を算出するステップと、
    　推定処理部が、算出された前記第１の分布および前記第２の分布に基づいて前記目標の個数および状態を観測時刻ごとに推定するステップと、
    　を備えたことを特徴とする目標推定方法。     Based on the first distribution, which is a probability distribution representing the number of targets in the observation range, and the second distribution, which is a distribution representing the density of the average number of targets in the state space representing the state of the target. A method of estimating the number and state of targets,
    The step of transforming the first distribution so that the distribution transforming portion keeps the variance of the distribution within the specified range.
    A step in which the prediction processing unit calculates the predicted value of the first distribution and the predicted value of the second distribution at the observation time by using the deformed first distribution and the second distribution. ,
    The update processing unit performs the first distribution at the observation time based on the predicted value of the first distribution at the observation time, the predicted value of the second distribution at the observation time, and the observation data of the motion specifications of the target at the observation time. Steps to calculate the distribution of 1 and the second distribution,
    A step in which the estimation processing unit estimates the number and state of the targets for each observation time based on the calculated first distribution and the second distribution.
    A target estimation method characterized by being equipped with.
PCT/JP2019/029768 2019-07-30 2019-07-30 Target estimation device and target estimation method WO2021019667A1 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
JP2021536503A JP6945778B2 (en) 2019-07-30 2019-07-30 Target estimation device and target estimation method
PCT/JP2019/029768 WO2021019667A1 (en) 2019-07-30 2019-07-30 Target estimation device and target estimation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2019/029768 WO2021019667A1 (en) 2019-07-30 2019-07-30 Target estimation device and target estimation method

Publications (1)

Publication Number Publication Date
WO2021019667A1 true WO2021019667A1 (en) 2021-02-04

Family

ID=74228385

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2019/029768 WO2021019667A1 (en) 2019-07-30 2019-07-30 Target estimation device and target estimation method

Country Status (2)

Country Link
JP (1) JP6945778B2 (en)
WO (1) WO2021019667A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022190719A1 (en) * 2021-03-09 2022-09-15 株式会社Soken Target recognition device

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110040525A1 (en) * 2009-08-17 2011-02-17 Lockheed Martin Corporation Method and system for calculating elementary symmetric functions of subsets of a set
JP2013522747A (en) * 2010-03-17 2013-06-13 アイシス イノベーション リミテッド How to track targets in video data
CN109525220A (en) * 2018-12-10 2019-03-26 中国人民解放军国防科技大学 Gaussian mixture CPHD filtering method with track association and extraction capability

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110040525A1 (en) * 2009-08-17 2011-02-17 Lockheed Martin Corporation Method and system for calculating elementary symmetric functions of subsets of a set
JP2013522747A (en) * 2010-03-17 2013-06-13 アイシス イノベーション リミテッド How to track targets in video data
CN109525220A (en) * 2018-12-10 2019-03-26 中国人民解放军国防科技大学 Gaussian mixture CPHD filtering method with track association and extraction capability

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
VO BA-TUONG ET AL.: "Analytic Implementations of the Cardinalized Probability Hypothesis Density Filter", IEEE TRANSACTIONS ON SIGNAL PROCESSING, vol. 55, no. 7, July 2007 (2007-07-01), pages 3553 - 3567, XP011185859, ISSN: 1053-587X, DOI: 10.1109/TSP.2007.894241 *
XUN FENG ET AL.: "A Multi-Target Track Before Detect Algorithm Based on CPHD for Pulse Doppler Radar", IEEE CONFERENCE PROCEEDINGS, 2016 FIRST IEEE ICCCI, vol. 10, 2016, pages 181 - 185, XP033019435, DOI: 10.1109/CCI. 2016.7778904 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022190719A1 (en) * 2021-03-09 2022-09-15 株式会社Soken Target recognition device
JPWO2022190719A1 (en) * 2021-03-09 2022-09-15
JP7448089B2 (en) 2021-03-09 2024-03-12 株式会社Soken Target recognition device

Also Published As

Publication number Publication date
JPWO2021019667A1 (en) 2021-10-21
JP6945778B2 (en) 2021-10-06

Similar Documents

Publication Publication Date Title
JP4849464B2 (en) Computerized method of tracking objects in a frame sequence
US20210232922A1 (en) Actor ensemble for continuous control
Niedfeldt et al. Multiple target tracking using recursive RANSAC
US10019801B2 (en) Image analysis system and method
EP2573734B1 (en) Systems and methods for evaluating plane similarity
US20170178357A1 (en) Moving object tracking apparatus, method and unmanned aerial vehicle using the same
CN109685830B (en) Target tracking method, device and equipment and computer storage medium
CN106600624B (en) Particle filter video target tracking method based on particle swarm
Zeng et al. Pixel modeling using histograms based on fuzzy partitions for dynamic background subtraction
WO2017168462A1 (en) An image processing device, an image processing method, and computer-readable recording medium
US20220080585A1 (en) Systems and methods for controlling a robot
CN105574892A (en) Doppler-based segmentation and optical flow in radar images
JP6945778B2 (en) Target estimation device and target estimation method
US11176702B2 (en) 3D image reconstruction processing apparatus, 3D image reconstruction processing method and computer-readable storage medium storing 3D image reconstruction processing program
KR101806453B1 (en) Moving object detecting apparatus for unmanned aerial vehicle collision avoidance and method thereof
CN109785372B (en) Basic matrix robust estimation method based on soft decision optimization
CN111968102A (en) Target equipment detection method, system, medium and electronic terminal
Baum et al. Extended object tracking based on combined set-theoretic and stochastic fusion
CN107273801B (en) Method for detecting abnormal points by video multi-target tracking
JP7211329B2 (en) Tracking device, tracking method, tracking program
CN104182971B (en) A kind of high precision image square localization method
WO2020053934A1 (en) Model parameter estimation device, state estimation system, and model parameter estimation method
US20180001821A1 (en) Environment perception using a surrounding monitoring system
CN109740470B (en) Target tracking method, computer device, and computer-readable storage medium
US20240220200A1 (en) Arithmetic operation processing device

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19939344

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2021536503

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19939344

Country of ref document: EP

Kind code of ref document: A1