Target motion analysis device, target motion analysis method, and target motion analysis program

Abstract

To solve the problem that a target motion analysis device capable of correctly estimating a state of a target without generating a sparse state, a target motion analysis method and a target motion analysis program are desired.SOLUTION: A target motion analysis device comprises: a weight update unit for updating a weight of a particle based on observation azimuth inputted at a predetermined time, state amounts of a plurality of particles disposed around the observation azimuth, and the weight of the particle; a state estimation unit for estimating a target state amount based on the state amounts of the particles and the updated weight of the particle; a motion update unit for calculating state amounts of particles at the next time based on the state amounts of the particles and the weight of the particle; a parameter estimation unit for estimating a target state amount based on observation azimuth at a plurality of times; and a particle position update unit for substituting the state amounts of the particles based on the estimated state amounts, the calculated state amounts of the particles and the weight of the particle. The particle position update unit substitutes a state amount of a particle with a minimum weight among the plurality of particles with the estimate state amount.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present invention relates to a target motion analysis device, a target motion analysis method, and a target motion analysis program for analyzing target motion from observed values in the fields of sonar and radar. [Background technology]

[0002] Conventionally, in the fields of sonar, radar, etc., target motion has been predicted from observed values. In order to predict target motion, it is necessary to estimate the target's position and velocity, and various methods have been proposed as a method for analyzing such target motion. For example, Patent Documents 1 to 4 disclose various target motion analysis methods using particle filters.

[0003] Target motion analysis using a particle filter is a method of representing the state of a target, including its position and velocity, using particles called "particles." In this target motion analysis, a large number of particles with various target states are first generated. Each particle has the same velocity and position state as the state of the target to be estimated, and processing is performed to increase particles that match the observed values and decrease particles that do not. Points with a high particle density indicate a high probability that the target is present, so the target's position and velocity can be estimated by gathering particles in places where the target is likely to be present. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Beyond the Kalman Filter Chapter 3 (pp. 35-65), Branko RISTIC, 2003 [Non-patent document 2] Thomas Brehard, Jean-Pierre Le Cadre, Initialization of Particle Filter and Posterior Cramer-Rao Bound for Bearings-Only Tracking in Modified Polar Coordinate System, [Research Report] RR-5074, INRIA, 2004 [Non-patent document 3] Thomas Brehard, Jean-Pierre Le Cadre, A New Approach for the Bearings-Only Problem: estimation of the variance-to-range ratio, 2004 [Non-patent document 4] Augustine Kong, Jun S. Liu and Wing Hung Wong, Sequential Imputations and Bayesian Missing Data Problems, Journal of the American Statistical Association, Vol. 89, No. 425 (Mar., 1994), pp.278-288, 1994 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in target motion analysis using a particle filter, if the target is located far away or if the estimation error of the target state is large due to a large observation error, a sparse state in which the number of particles around the target is small may occur. Also, a sparse state may occur due to a bias error in the observation direction.

[0006] When a sparsely populated state occurs, the number of particles present in the vicinity of the target's observation direction decreases, making it difficult to accurately estimate the target's state. Furthermore, once a sparsely populated state occurs, the target's state cannot be accurately estimated in subsequent processing. Therefore, there is a demand for a target motion analysis device, a target motion analysis method, and a target motion analysis program that can accurately estimate the target's state without causing a sparsely populated state. [Means for solving the problem]

[0007] The target motion analysis device according to the present invention is a target motion analysis device that estimates a state of a target based on an observation direction indicating the direction of the target, and state quantities of a plurality of particles arranged around the observation direction and weights of the particles, and includes a weight update unit that updates the weights of the particles based on the observation direction, state quantities of the particles, and weights of the particles input at a predetermined time, and a weight update unit that updates the state quantities of the particles and the updated weights of the particles. from , the target 1st Estimate the state quantity and output as the estimation result. and a state estimation unit for estimating the particle state based on the particle state quantity and the particle weight at the next time. The above used for estimation a motion update unit that calculates the state quantities of the particles; before the predetermined time The observation directions at multiple times from , the target Second The state quantity As an estimated state quantity a parameter estimation unit for estimating the Estimated state quantity and based on the state quantities of the particles and the weights of the particles calculated by the motion update unit, The motion update unit calculates a particle position update unit that replaces the state quantity of the particle, and the particle position update unit is configured to update the particle position of the particle having the smallest weight among the plurality of particles; The aforementioned The state quantity of the particle is replaced with the estimated state quantity estimated by the parameter estimation unit.

[0008] Furthermore, a target motion analysis method according to the present invention is a target motion analysis method for estimating a state of a target based on an observation direction indicating the direction of the target, and state quantities of a plurality of particles arranged around the observation direction and weights of the particles, the method including a weight updating step of updating the weights of the particles based on the observation direction input at a predetermined time, the state quantities of the particles, and the weights of the particles; from , the target 1st Estimate the state quantity and output as the estimation result. and a state estimation step of estimating the particle state based on the particle state and the particle weight at the next time. The above used for estimation a motion update step of calculating the state quantities of the particles; before the predetermined time The observation directions at multiple times from , the target Second The state quantity As an estimated state quantity a parameter estimation step for estimating the Estimated state quantity and based on the state quantities of the particles and the weights of the particles calculated in the motion update step, The motion update step a particle position updating step of replacing the state quantities of the particles, wherein in the particle position updating step, the particle position is updated so that the particle position is updated to the particle position having the smallest weight among the plurality of particles. The aforementioned The state quantities of the particles are replaced with the estimated state quantities estimated in the parameter estimation step.

[0009] The target motion analysis program according to the present invention is a target motion analysis program that estimates a state of a target based on an observation direction indicating the direction of the target, and state quantities of a plurality of particles arranged around the observation direction and weights of the particles, and includes a weight updating step of updating the weights of the particles based on the observation direction, state quantities of the particles, and weights of the particles input at a predetermined time, and a weight updating step of updating the state quantities of the particles and weights of the particles. from , the target 1st Estimate the state quantity and output as the estimation result. and a state estimation step of estimating the particle state based on the particle state and the particle weight at the next time. The above used for estimation a motion update step of calculating the state quantities of the particles; before the predetermined timeThe observation directions at multiple times from , the target Second The state quantity As an estimated state quantity a parameter estimation step for estimating the Estimated state quantity and based on the state quantities of the particles and the weights of the particles calculated in the motion update step, The motion update step a particle position update step of replacing the state quantity of the particle by a processor of the target motion analysis device, and The aforementioned The state quantities of the particles are replaced with the estimated state quantities estimated in the parameter estimation step. [Effects of the Invention]

[0010] As described above, according to the present invention, the parameter estimation unit estimates the state quantity of the target, and the particle position update unit replaces the state quantity of the particle with the smallest weight among the plurality of particles with the estimated state quantity. This keeps the weight of the particle located near the observation direction large, so that the state of the target can be accurately estimated without causing a sparse state. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 is a block diagram showing an example of the configuration of a conventional target motion analysis device. [Figure 2] 10 is a schematic diagram for explaining a weight update process performed by a weight update unit. FIG. [Figure 3] FIG. 10 is a schematic diagram for explaining a resampling process performed by a resampling unit. [Figure 4] FIG. 10 is a schematic diagram for explaining a weight update process by a weight update unit in a depopulated state. [Figure 5] FIG. 10 is a schematic diagram for explaining the resampling process by the resampling unit in a depopulated state. [Figure 6] 1 is a block diagram showing an example of the configuration of a desired motion analysis device according to a first embodiment. [Figure 7]10 is a schematic diagram for explaining a replacement process performed by a particle position update unit. FIG. [Figure 8] FIG. 4 is a schematic diagram for explaining a weight update process by a weight update unit in a depopulated state in the first embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, embodiments of the present invention will be described with reference to the drawings. The present invention is not limited to the following embodiments, and various modifications are possible without departing from the spirit of the present invention. Furthermore, the present invention includes all possible combinations of the configurations shown in the following embodiments. In addition, in each drawing, components with the same reference numerals are the same or equivalent, and this is common throughout the entire specification.

[0013] Embodiment 1 A description will be given of a target motion analysis device according to the present embodiment 1. The target motion analysis device according to the present embodiment 1 analyzes the motion of a target in order to predict the state of the target from observation values including information indicating the direction of the target observed by sonar, radar, or the like.

[0014] [About conventional target motion analysis devices] First, a conventional target motion analysis device will be described before describing the target motion analysis device according to the present embodiment 1. As described in the background art section, in a target motion analysis device using a particle filter, a large number of particles having various target states are generated, and target motion analysis processing is performed so as to increase particles that match the observed values and reduce particles that do not match the observed values.

[0015] (modified polar coordinates) State quantity X of each particle k (i) When expressed in Cartesian coordinates, is defined as in equation (1). In equation (1), "x" indicates the x-axis coordinate of the particle's position, and "y" indicates the y-axis coordinate of the particle's position. "v x" indicates the velocity in the x-axis direction, and "v y " indicates the velocity in the y-axis direction. "k" is an index indicating time, and "i" is an index indicating the particle number. Furthermore, "N" indicates the number of particles.

[0016]

number

[0017] Each particle has a weight that indicates the reliability of the particle. A large weight indicates that the state quantity indicated by the particle is likely to be the target state quantity to be estimated.

[0018] However, when the observation only indicates the direction, the target's position and velocity are not directly observed. In particular, before the observer changes course, there is no information about the distance, and in this case, all state quantities, including the target's position and velocity, are indeterminate.

[0019] In such cases, modified polar coordinates are used, which are a coordinate system that can separate variables that can be estimated from the observed value of only the orientation from variables that cannot be estimated. The state quantity Y k (i) is expressed by equation (2). In equation (2), "β" indicates the direction and "r" indicates the distance. "β with a dot on it" indicates the direction change, and "r / r with a dot on it" indicates the distance change rate.

[0020]

number

[0021] Using the modified polar coordinates shown in equation (2), even before the observer changes course, 1,k (i) " ~ "Y 3,k (i) " can be calculated from the observation direction. 4,k (i)" can be determined by the observer changing course.

[0022] Here, the relationship between Cartesian coordinates and modified polar coordinates is expressed by equations (3) and (4). In equation (3), "f c mp " represents a conversion function for converting from Cartesian coordinates to modified polar coordinates. Also, in equation (4), "f mp c " denotes the conversion function for converting from modified polar coordinates to Cartesian coordinates.

[0023]

number

[0024]

number

[0025] (Conventional target motion analysis device 100) Fig. 1 is a block diagram showing an example of the configuration of a conventional target motion analysis device. Conventional target motion analysis device 100 performs target motion analysis processing to estimate the state of a target based on the target observation direction input to input terminal 20, and outputs the estimation result from output terminal 30. As shown in Fig. 1, target motion analysis device 100 includes particle initialization unit 11, weight update unit 12, state estimation unit 13, degeneracy determination unit 14, resampling unit 15, and motion update unit 16.

[0026] The particle initialization unit 11 performs initialization processing to uniformly arrange each particle in a predetermined range on the modified polar coordinate system, and outputs an initial state quantity, which is a state quantity of the initialized i-th particle, and an initial weight, which is a weight of the initial state quantity of the particle. The initialization processing by the particle initialization unit 11 is performed only once when the target motion analysis processing is started.

[0027] The weight update unit 12 calculates likelihood based on the observation direction indicating the direction of the target input to the input terminal 20 at a predetermined time, and the state quantities and weights of the particles generated by the motion update unit 16 or the particle initialization unit 11. In response to this, the weight update unit 12 performs a weight update process to update the weights of the particles. The weight update unit 12 also normalizes the updated weights of the particles.

[0028] The state estimation unit 13 performs a state estimation process to estimate the state quantities of the target on the modified polar coordinates based on the state quantities of the particles and the normalized weights output from the weight update unit 12. The state estimation unit 13 also converts the estimated state quantities, which are the estimation results on the modified polar coordinates, into estimated state quantities, which are the estimation results on the Cartesian coordinates.

[0029] The degeneracy determination unit 14 determines degeneracy based on the particle weights output from the weight update unit 12. The degeneracy determination unit 14 calculates a determination value for determining degeneracy, compares the calculated determination value with a preset threshold, and performs particle degeneracy determination processing. Here, "degeneracy" means that large weights are concentrated on a small number of particles, and the weights of the majority of other particles become very small.

[0030] When the degeneracy determination unit 14 determines that degeneracy has occurred, the resampling unit 15 performs a resampling process of resampling the state quantity and the weight of the particle based on the weight of the particle output from the degeneracy determination unit 14. The details of the resampling process will be described later.

[0031] The motion update unit 16 performs a motion update process to calculate the particle state quantities at the next time based on the particle state quantities and normalized particle weights output from the degeneracy determination unit 14 or the resampling unit 15. Then, the motion update unit 16 outputs the calculated particle state quantities at the next time to the weight update unit 12.

[0032] (Conventional target motion analysis processing) The target motion analysis process performed by the conventional target motion analysis device 100 having the above configuration will now be described. When the target motion analysis process is started, first, the particle initialization unit 11 uniformly arranges each particle within a predetermined range on the modified polar coordinate system, as described in Non-Patent Document 2. Here, at the time of initialization, there is no knowledge about the particles, so the particle initialization unit 11 uniformly weights each particle.

[0033] The initial state quantity Y1 is the state quantity of the initialized i-th particle. (i) , and the initial state quantity Y1 of the i-th particle (i) The initial weight w1 is the weight of (i) is output to the weight update unit 12. Note that the processing by the particle initialization unit 11 is executed only once at the start of the processing.

[0034] The weight update unit 12 calculates likelihood using the observation direction input to the input terminal 20 at a predetermined time, and the state quantities and weights of the particles generated by the motion update unit 16 or the particle initialization unit 11, and performs weight update processing to update the weights of the particles.

[0035] FIG. 2 is a schematic diagram for explaining the weight update process by the weight update unit. FIG. 2(a) shows likelihoods calculated based on the observation direction. The direction with the largest calculated likelihood value becomes the observation direction indicating the direction in which the target is observed. FIG. 2(b) shows particle weights output from the movement update unit 16. Normally, particle weights output from the movement update unit 16 have large values near the observation direction. FIG. 2(c) shows particle weights updated by the weight update unit 12. The weight update unit 12 performs weight update process to update the particle weights as shown in FIG. 2(c) by multiplying the likelihoods shown in FIG. 2(a) by the particle weights shown in FIG. 2(b).

[0036] Specifically, the updated particle weights w k (i) is calculated based on formula (5). In formula (5), "z k " indicates the observation direction, and "p(z k |Y 3,k (i))" indicates the likelihood function. k-1 (i) " indicates the weight of the particle at the previous time index.

[0037]

number

[0038] After the weight update process for each particle is completed, the weight update unit 12 normalizes the updated weight of each particle. k (i) is performed based on equation (6). In the following explanation, a letter such as "w" followed by "~" in a mathematical formula will be written as "w~".

[0039]

number

[0040] The state quantity Y of each particle k (i) , and the normalized particle weights w~ k (i) are output to the state estimation unit 13 and the degeneracy determination unit 14. At this time, the state quantities of the particles generated by the particle initialization unit 11 or the motion update unit 16 are output to the state estimation unit 13 and the degeneracy determination unit 14 while being retained.

[0041] The state estimation unit 13 calculates the particle state quantity Y k (i) and the normalized weight w~ k (i) Based on this, a state estimation process is performed to estimate the state quantity of the target on the modified polar coordinates, and the estimated state quantity Y^ is obtained as the estimation result. k The estimated state quantity Y^ of the target on the modified polar coordinate system, which is the estimation result, is output. k As described in Non-Patent Document 3, the weighted sum is calculated based on Equation (7).

[0042]

number

[0043] Then, the state estimation unit 13 calculates the estimated state quantity Y^ of the target on the corrected polar coordinates, which is the estimation result, based on the equation (4). k is the estimated state quantity X^ of the target on Cartesian coordinates. k The state estimation unit 13 converts the estimated state quantity X^ of the target on the Cartesian coordinate system into k is output as the estimation result.

[0044] The degeneracy determination unit 14 determines the particle weights w~ output from the weight update unit 12. k (i) Furthermore, as described in Non-Patent Document 4, the degeneration determination unit 14 calculates the determination value based on the formula (8) and a preset threshold value N threshold and performs a degeneracy determination process for the particle.

[0045]

number

[0046] If the judgment value is smaller than the threshold, that is, if the relationship in equation (8) holds, the degeneracy judgment unit 14 judges that degeneracy has occurred. In this case, the state quantities and weights of the particles are output to the resampling unit 15. On the other hand, if the judgment value is equal to or larger than the threshold, that is, if the relationship in equation (8) does not hold, the degeneracy judgment unit 14 judges that degeneracy has not occurred. In this case, the state quantities and weights of the particles are output to the motion update unit 16.

[0047] If the degeneracy determination unit 14 determines that degeneracy has occurred, the resampling unit 15 performs a resampling process to resample the particle state quantities and particle weights output from the degeneracy determination unit 14. The particle state quantities and particle weights obtained by the resampling process are output to the motion update unit 16.

[0048] 3A and 3B are schematic diagrams for explaining the resampling process by the resampling unit. Fig. 3A shows the weights of the particles output from the degeneracy determination unit 14. As shown in Fig. 3A, the particles are present in the vicinity of the observation direction, and degeneracy has occurred.

[0049] Fig. 3(b) shows the positions of the resampled particles. As shown in Fig. 3(b), the resampling unit 15 deletes particles with small weights (white circles indicated by dashed lines) based on the particle weights output from the degeneracy determination unit 14. The resampling unit 15 also divides particles with large weights according to the magnitude of the particle weights, and replicates the particles (white circles indicated by solid lines).

[0050] When dividing and duplicating particles, the resampling unit 15 calculates a cumulative distribution function from the weights output from the degeneracy determination unit 14, and assigns particles by resampling the calculated cumulative distribution function at equal intervals for the number of particles. At this time, the resampling unit 15 equalizes the weights of the assigned particles. This essentially results in the particles with small weights being deleted, and the particles with large weights being divided and duplicated.

[0051] 3(c) shows how the particles are gathered around the observation direction by resampling, and the particle weights are uniform. The resampled particle state quantities and particle weights are output to the motion update unit 16.

[0052] The motion update unit 16 calculates the particle state quantity Y1 output from the degeneracy determination unit 14 or the resampling unit 15. (i) and the normalized weights w~ k (i) The state quantity of the particle at the next time is calculated in accordance with the motion model of the target based on the above. The calculated state quantity of the particle is output to the weight update unit 12.

[0053] (Problems with the conventional target motion analysis device 100) As explained in the background art section, when the target motion analysis device 100 using the above algorithm has a large estimation error in the state of the target, or due to a bias error in the observation direction, a depopulated state may occur in which the number of particles around the state of the target is small or there is no particle at all.

[0054] Fig. 4 is a schematic diagram for explaining the weight update process by the weight update unit in a depopulated state. Fig. 4(a) shows the likelihood calculated based on the observation direction. Fig. 4(b) shows the particle weights output from the movement update unit 16. Fig. 4(c) shows the particle weights updated by the weight update unit 12.

[0055] In this example, since there are no particles near the observation direction, the weights of the particles output from the weight update unit 12 are also small, as shown in Fig. 4(c). As a result, it becomes difficult for the conventional target motion analysis device 100 to correctly estimate the state of the target. Furthermore, since there are no particles with large weights, even if the resampling process is performed by the resampling unit 15, the particles do not gather around the observation direction.

[0056] Fig. 5 is a schematic diagram for explaining the resampling process by the resampling unit in a sparsely populated state. Fig. 5(a) shows the particle weights output from the degeneracy determination unit 14. Fig. 5(b) shows the positions of the resampled particles. Fig. 5(c) shows how the particles are biased to positions away from the observation direction due to resampling, resulting in uneven particle weights.

[0057] As shown in the example of FIG. 5, when there are no particles in the vicinity of the observation direction, the weight output from the degeneracy determination unit 14 becomes small. Therefore, even if particles with small weights are deleted in the resampling process by the resampling unit 15 and particles with large weights are replicated, particles will not gather around the observation direction. Furthermore, in some cases, once a depopulation state occurs, it may never be possible to correctly estimate the state of the target again. Therefore, in the first embodiment, processing is performed to suppress a decrease in the accuracy of target state estimation due to the occurrence of a particle depopulation state.

[0058] [Configuration of target motion analysis device] 6 is a block diagram showing an example of the configuration of a target motion analysis device according to the first embodiment. As shown in FIG. 6, the target motion analysis device 10 includes a particle initialization unit 11, a weight update unit 12, a state estimation unit 13, a degeneracy determination unit 14, a resampling unit 15, a motion update unit 16, a parameter estimation unit 17, and a particle position update unit 18. Such a target motion analysis device 10 is configured by a calculation device such as a microcomputer that realizes various functions by executing software, or hardware such as a circuit device corresponding to the various functions. Note that in the target motion analysis device 10 according to the first embodiment, the configurations of the particle initialization unit 11, the state estimation unit 13, the degeneracy determination unit 14, and the resampling unit 15 are common to those of the conventional target motion analysis device 100. Therefore, detailed description thereof will be omitted here.

[0059] The weight update unit 12 calculates likelihood based on the observation direction input to the input terminal 20 and the state quantities and weights of the particles generated by the particle position update unit 18 or the particle initialization unit 11. In response to this, the weight update unit 12 performs a weight update process to update the weights of the particles. The weight update unit 12 also normalizes the updated weights of the particles.

[0060] The motion update unit 16 performs a motion update process to calculate the state quantity of the particle at the next time based on the state quantity of the particle and the normalized particle weight output from the degeneracy determination unit 14 or the resampling unit 15. Then, the motion update unit 16 outputs the calculated state quantity of the particle at the next time to the particle position update unit 18.

[0061] The parameter estimation unit 17 receives observation directions at a plurality of times from an input terminal 20 and estimates the state quantities of the target based on the input observation directions at the plurality of times. The parameter estimation unit 17 outputs the estimated state quantities of the target as estimated state quantities to the particle position update unit 18. In this way, the parameter estimation unit 17 estimates the state quantities of the target based only on the observation directions at a plurality of times without using particles.

[0062] The particle position update unit 18 performs a replacement process to replace the particle state quantities based on the estimated state quantities output from the parameter estimation unit 17 and the particle state quantities and particle weights output from the motion update unit 16, and updates the particle state quantities and particle weights. The particle position update unit 18 outputs the updated particle state quantities and particle weights to the weight update unit 12.

[0063] (Target motion analysis processing) The following describes the target motion analysis processing performed by the target motion analysis device 10 having the above configuration. In this target motion analysis processing, the particle initialization unit 11, state estimation unit 13, degeneracy determination unit 14, resampling unit 15, and motion update unit 16 perform the same processing as in the target motion analysis processing performed by the conventional target motion analysis device 100, and therefore detailed description thereof will be omitted here.

[0064] The parameter estimation unit 17 holds a plurality of observation directions input to an input terminal 20, and calculates the state quantity Y'^ of the target from the held plurality of observation directions. k Estimate the target state Y´^ k For example, the method described in Non-Patent Document 2 can be used to estimate Z. This method is a method of linearly approximating the non-linear observation equation shown in equation (9) and solving the equation.t " indicates the observation direction. "t" is an index indicating the time of the obtained observation direction, and "k" is an index indicating the current time. "δ t ” indicates the observation time interval, and “σ v " indicates the observation error. Also, "N(0, σ v 2 ) has a mean of 0 and a variance of σ v 2 It is a random number that follows a normal distribution where , and indicates Gaussian noise.

[0065]

number

[0066] In the above equation (9), "tan -1 (sinθ / cosθ)=θ”, and assuming that the orientation change is linear within multiple held times, by setting “cosθ=1” and “sinθ=θ”, the state quantity Y 1,k ~Y 3,k can be solved as shown in equation (10).

[0067]

number

[0068] Furthermore, "Z" and "X" in formula (10) are calculated based on formula (11). k " indicates the estimated state quantity, which is the estimated value of the state quantity, "Z" indicates the observation direction, and "δ t " indicates the observation time interval. The estimated state quantity Y'^ of the target estimated in this way k is output to the particle position update unit 18.

[0069]

number

[0070] The particle position update unit 18 updates the state quantities of one or more particles with a small weight among the state quantities of the particles output from the motion update unit 16 by the estimated state quantity Y'^ output from the parameter estimation unit 17. k The particle position update unit 18 performs a replacement process to replace the estimated state quantity Y'^ k The state quantity Y including the particle state quantity replaced by k (i) and the weights of the particle state quantities to the weight update unit 12.

[0071] FIG. 7 is a schematic diagram for explaining the replacement process by the particle position update unit. Here, a case will be explained in which the state quantity of one particle with the smallest weight is replaced with an estimated state quantity. FIG. 7(a) shows the particle weights output from the motion update unit 16. In this example, it is shown that no particles exist near the observation direction. FIG. 7(b) shows the weights output from the particle position update unit 18. In FIG. 7(b), the target state quantity of the particle with the smallest weight among the particles output from the motion update unit 16 is the estimated state quantity Y'^ calculated by the parameter estimation unit 17. k At this time, the weight of the replaced particle is reset to, for example, "1".

[0072] The weight update unit 12 calculates likelihoods using the observation orientation input to the input terminal 20, and the state quantities and weights of the particles generated by the particle position update unit 18 or the particle initialization unit 11, and performs weight update processing to update the weights of the particles. In addition, the weight update unit 12 normalizes the updated weights of each particle, and calculates the normalized weights w~ of each particle. k (i) Output.

[0073] Thereafter, the state estimation unit 13, the degeneracy determination unit 14, the resampling unit 15, and the motion update unit 16 perform processing for each model in the same manner as the target motion analysis processing by the conventional target motion analysis device 100. As a result, the state estimation unit 13 outputs the estimated state quantity X^ of the target on the Cartesian coordinate system. k is output as the estimation result.

[0074] In this way, by adding the parameter estimation unit 17 and the particle position update unit 18 to the conventional target motion analysis device 100, when the weight of the particle is updated in the weight update unit 12, the particle is positioned in a direction with a high likelihood.

[0075] Fig. 8 is a schematic diagram for explaining the weight update process by the weight update unit in a depopulated state in the first embodiment. Fig. 8(a) shows the likelihood calculated based on the observation direction. Fig. 8(b) shows the particle weights output from the particle position update unit 18. Fig. 8(c) shows the particle weights updated by the weight update unit 12.

[0076] The conventional target motion analysis device 100 shown in FIG. 4 has a problem in that when there are few particles near the observation direction, the weight of the updated particles also becomes small. In contrast, the target motion analysis device 10 according to the first embodiment shown in FIG. 8 replaces particles with small weights with particles estimated by the parameter estimation unit 17. This allows particles to be placed in positions with high likelihood, and the weight of particles placed near the observation direction can be maintained at a high level. As a result, particles are re-gathered around the observation direction by resampling. This prevents a depopulation state and allows the state of the target to be accurately estimated.

[0077] As described above, in the target motion analysis device 10 according to the first embodiment, the parameter estimation unit 17 estimates the state quantity of the target based on the observation direction at a plurality of times, and the particle position update unit 18 replaces the state quantity of the particle with the smallest weight among the plurality of particles with the state quantity of the target estimated by the parameter estimation unit 17. This keeps the weight of the particles arranged near the observation direction large, preventing a sparse state and enabling the state of the target to be correctly estimated.

[0078] Although the first embodiment of the present invention has been described above, the present invention is not limited to the first embodiment described above, and various modifications and applications are possible within the scope of the present invention. For example, in the first embodiment, the parameter estimator 17 estimates the target state variable Y'^k Although the method of estimating is described above as applying a method of solving the nonlinear equation of equation (9) by linear approximation, the present invention is not limited to this example. For example, the target state Y´^ k As a method for estimating , a method using a Kalman filter described in a conventional document (Asano Futoshi, Array Signal Processing of Sound (pp. 171-179), Corona Publishing, 2011) may be applied, or the Gauss-Newton method described in Non-Patent Document 2 may be applied. Even when these estimation methods are applied, the same effect as in the first embodiment can be obtained.

[0079] In the first embodiment, the particle position update unit 18 uses only the state quantity of one particle with a small weight among the state quantities of other particles obtained from the motion update unit 16 as the estimated state quantity Y'^ k However, the present invention is not limited to this, and multiple state quantities of particles may be replaced. By replacing multiple state quantities of particles, it is thought that particles are more likely to gather near the observation direction when updating the particle weights. However, if too many particles are replaced, degeneration may occur, which may worsen the estimation result.

[0080] Furthermore, in the first embodiment, the particle position update unit 18 resets the particle weight to the value "1" in order to increase the weight of particles in the vicinity of the observation direction. However, this is not limited to this, and the weight may be reset to a value other than "1", such as 1 times the number of particles. [Explanation of symbols]

[0081] 10, 100 target motion analysis device, 11 particle initialization unit, 12 weight update unit, 13 state estimation unit, 14 degeneracy determination unit, 15 resampling unit, 16 motion update unit, 17 parameter estimation unit, 18 particle position update unit, 20 input terminal, 30 output terminal.

Claims

1. A target motion analysis device that estimates a state of a target based on an observation direction indicating a direction of the target, and state quantities of a plurality of particles arranged around the observation direction and weights of the particles, comprising: a weight update unit that updates the weights of the particles based on the observation direction input at a predetermined time, the state quantities of the particles, and the weights of the particles; a state estimation unit that estimates a first state quantity of the target from the state quantities of the particles and the updated weights of the particles, and outputs the estimated result; a motion update unit that calculates a state quantity of the particle to be used for estimation at a next time based on the state quantity of the particle and the weight of the particle; a parameter estimation unit that estimates a second state quantity of the target as an estimated state quantity from the observation azimuths at a plurality of times before the predetermined time; a particle position update unit that replaces the state quantity of the particle calculated by the motion update unit based on the estimated state quantity estimated by the parameter estimation unit and the state quantity of the particle and the weight of the particle calculated by the motion update unit; Equipped with The particle position update unit The state quantity of the particle with the smallest weight among the plurality of particles is replaced with the estimated state quantity estimated by the parameter estimation unit. Target motion analysis device.

2. a degeneracy determination unit that receives the state quantities of the particles and the weights of the particles output from the weight update unit and determines whether degeneracy has occurred based on the weights of the particles updated by the weight update unit; a resampling unit that, when it is determined that the degeneration has occurred, resamples the state quantities of the particles and the weights of the particles based on the weights of the particles output from the degeneration determination unit; Furthermore, The motion update unit Calculating the state quantity of the particle at the next time based on the state quantity of the particle and the weight of the particle output from the degeneracy determination unit or the resampling unit The target motion analysis device according to claim 1 .

3. The degeneration determination unit calculating a judgment value for judging the degeneracy based on the input weight of the particle; The calculated determination value is compared with a preset threshold value to determine whether the degeneration has occurred. The target motion analysis device according to claim 2 .

4. The resampling unit When it is determined that the degeneration has occurred, the particles with the small weights are deleted, and the particles with the large weights are divided and replicated, thereby performing the resampling.

4. The target motion analysis device according to claim 2 or 3.

5. The particle initialization unit further includes, when the observation direction is first input, a particle initialization unit that uniformly arranges the plurality of particles having initial state quantities and initial weights around the observation direction. The target motion analysis device according to any one of claims 1 to 4.

6. A target motion analysis method for estimating a state of a target based on an observation direction indicating a direction of the target, and state quantities of a plurality of particles arranged around the observation direction and weights of the particles, comprising: a weight updating step of updating the weights of the particles based on the observation orientations input at predetermined times, the state quantities of the particles, and the weights of the particles; a state estimation step of estimating a first state quantity of the target from the state quantities of the particles and the updated weights of the particles, and outputting the result as an estimation result; a motion update step of calculating a state quantity of the particle to be used for estimation at a next time based on the state quantity of the particle and the weight of the particle; a parameter estimation step of estimating a second state quantity of the target as an estimated state quantity from the observation azimuths at a plurality of times before the predetermined time; a particle position update step of replacing the state quantity of the particle calculated in the motion update step based on the estimated state quantity estimated in the parameter estimation step, and the state quantity of the particle and the weight of the particle calculated in the motion update step; Equipped with In the particle position updating step, The state quantity of the particle with the smallest weight among the plurality of particles is replaced with the estimated state quantity estimated in the parameter estimation step. Target motion analysis method.

7. A target motion analysis program that estimates a state of a target based on an observation direction indicating a direction of the target, state quantities of a plurality of particles arranged around the observation direction, and weights of the particles, comprising: a weight updating step of updating the weights of the particles based on the observation orientations input at predetermined times, the state quantities of the particles, and the weights of the particles; a state estimation step of estimating a first state quantity of the target from the state quantities of the particles and the updated weights of the particles, and outputting the result as an estimation result; a motion update step of calculating a state quantity of the particle to be used for estimation at a next time based on the state quantity of the particle and the weight of the particle; a parameter estimation step of estimating a second state quantity of the target as an estimated state quantity from the observation azimuths at a plurality of times before the predetermined time; a particle position update step of replacing the state quantity of the particle calculated in the motion update step based on the estimated state quantity estimated in the parameter estimation step, and the state quantity of the particle and the weight of the particle calculated in the motion update step; causing a processor of the target motion analysis device to execute the above; In the particle position updating step, The state quantity of the particle with the smallest weight among the plurality of particles is replaced with the estimated state quantity estimated in the parameter estimation step. Targeted motion analysis program.

Images