JP5203654B2 - Target motion analysis method and apparatus - Google Patents

Description

  The present invention relates to a target motion analysis method and apparatus for observing sound waves radiated from a target body as time-series data and estimating a target position, course, speed, and the like.

  The target motion analysis method uses an acoustic sensor provided on a movable observation body to accumulate the arrival direction and frequency of sound waves radiated from the target body as time series observation data, and the time series observation data and time series estimation data The parameter of the motion model of the target body that minimizes the residual is estimated.

  Conventional target motion analysis methods treat a target body motion model as a constant-velocity linear motion, and parameters to be estimated are the position coordinates of the target body at the reference time, the course, the speed, and the natural frequency (Patent Document 1). .

  The problem of minimizing the residual between the time series observation data and the time series estimation data and estimating the motion vector of the target body is known as a nonlinear optimization problem, and is based on the Kalman filter method, the least square method, the genetic There was a method based on genetic algorithm.

  Further, there has been a method for detecting a change in shift and presenting it to an operator for a target that is not a constant-velocity linear movement, that is, a target that performs a change in needle shift (Patent Document 2).

JP-A-10-62511 JP-A-10-62507

  In the above-described technique, the motion model of the target body is handled as a constant-velocity linear motion, and therefore there is a problem that if the target body performs a needle changing speed change, the target motion before and after the needle changing speed cannot be analyzed correctly.

  FIG. 11 shows a simulation result of the target motion analysis according to the above-described prior art. 1101 is the track of the observation object, 1102 is the track of the target object, 1103 is the estimated position coordinate history of the target object when the target motion analysis is performed with an analysis period of 30 seconds, 1104 is the relative distance between the observation object and the target object, 1105 is Estimated relative distance between the observation object and the target body based on the target motion analysis result, 1106 is the target body speed, 1107 is the estimated target speed based on the target motion analysis result, 1108 is the course of the target body, 1109 is the target based on the target motion analysis result Represents the estimated course of the body. In 900 seconds after the simulation starts, the target body changes its course from 180 ° to 150 °, and it can be seen that the target motion analysis cannot be performed correctly after the target body change.

  Also, in the method of analyzing the target body speed change by some verification means and using the time-series data after the speed change speed change, the analysis becomes impossible immediately after the speed change speed change, and the analysis result becomes unstable. There was a problem. This is because the number of data used for analysis is reduced after the gear change.

  FIG. 12 shows a simulation result of target motion analysis by a technique for detecting a target shift by some verification means, and analyzing the target motion using observation data after the shift change time after detecting the shift change. 1201 is the track of the observation object, 1202 is the track of the target object, 1203 is the estimated position coordinate history of the target object when the target motion analysis is performed with an analysis period of 30 seconds, 1204 is the relative distance between the observation object and the target object, 1205 is Estimated relative distance between the observation object and the target body based on the target motion analysis result, 1206 is the target body speed, 1207 is the target body speed estimated from the target motion analysis result, 1208 is the course of the target body, 1209 is the target based on the target motion analysis result Represents the estimated course of the body. 900 seconds after the simulation starts, the target has changed its course from 180 ° to 150 °. After the target has changed, the analysis is once impossible and the analysis result is unstable for about 600 seconds. I understand.

  As a method for solving the above-mentioned problems, there is a method for modeling the target body motion as a combination of two or more constant speed straight ahead sections in consideration of variable speed shifts instead of constant speed straight forward motion as a target body motion model. It is conceivable (hereinafter, this model is referred to as a variable speed transmission model with respect to the constant speed linear model). However, the time series estimation azimuth function and time series frequency function of the variable speed transmission model become discontinuous, and it is possible to apply an optimization method that uses the derivative of the evaluation value, such as the Kalman filter or the least square method. Can not. Although it is relatively easy to apply the variable speed model to a genetic algorithm, it is known that the amount of calculation for converging the analysis dramatically increases as the estimated parameters of the genetic algorithm increase.

  Therefore, there is a need for a target motion analysis method and apparatus that stably outputs a motion vector of a target body before and after a gear change without significantly increasing the amount of calculation.

  The present invention provides a target motion analysis that estimates the position, course, and speed of a target body from time-series observation direction data and time-series observation frequency data obtained by observing the arrival direction and frequency of sound waves radiated from the target body with an acoustic sensor. Is the method. The target motion analysis method estimates the target shift speed after detecting the target shift of the target body, and uses the time-series observation data to determine the target body position coordinates at the reference time and the target body position for each constant speed straight section. It is to solve the nonlinear optimization problem using the course and speed as estimation parameters.

  More specifically, first, assuming the time of the target changing gear shift, it is assumed that each of the sections before and after the assumed changing gear shift time is moving at a constant speed, and the target moves straight at a constant speed. For each motion section, obtain time-series estimated azimuth data based on time-series observation azimuth data and time-series estimated frequency data based on time-series observation frequency data. And the sum of the residuals of the time-series observed frequency data and the time-series estimated frequency data are used as the evaluation value to estimate the time of the needle changing speed. Considering each of the sections before and after the estimated time of the variable speed shift, the target body is moving straight ahead at a constant speed, and for each constant speed straight movement section, time series estimation based on time series observation azimuth data is performed. The time series estimated frequency data based on the azimuth data and the time series observation frequency data is obtained, the residual between the time series observation direction data and the time series estimation direction data, and the time series observation frequency data and the time series estimation frequency data The motion vector of the target body is estimated using the sum of residuals as an evaluation value.

  With the target motion analysis method and apparatus of the present invention, the motion vector of the target body can be stably estimated even for a target body that performs a zigzag motion that repeats variable speed shifting.

  Embodiments of a target motion analysis method and apparatus according to the present invention will be described below with reference to the drawings.

  FIG. 1 is a functional block diagram for realizing the target motion analysis method. The change-of-needle shift detection unit 102 refers to the observation information database 101 and the change-of-needle shift database 103, detects whether or not a new change of needle change has occurred, and estimates the change-of-needle shift time. The target motion analysis unit 104 refers to the observation information database 101 and the needle changing speed change database 103, performs an analysis to estimate the target body position at the reference time, the course and speed at each constant speed, and stores the analysis result in the analysis history database 105. To do. When the needle shift is not registered on the needle shift database 103, an analysis for estimating the target position and speed at the reference time is performed, and the analysis result is stored in the analysis result database 105.

  FIG. 2 is a diagram for explaining the observation information database unit 101. As shown in the figure, the observation direction database and the observation frequency database are configured independently. [] Indicates an array structure for accumulating data in time series, and the number of accumulated data is stored as the number of stored data. Note that the observation azimuth error standard deviation and observation frequency error standard deviation indicate the accuracy of the observed information and are not necessarily required information, but it is possible to analyze with higher accuracy by using this information. It becomes.

  FIG. 3 is a diagram for explaining the structure of the variable speed transmission database 103. It has a structure that has a record that stores the estimated time for changing the needle changing speed as the speed changing information, and a record that holds the number of times of changing the needle changing speed.

  FIG. 4 is a flowchart for explaining the processing of the variable speed shift detection unit 102. In the process of 401, the latest shift change time is acquired from the shift change database 103, and the data after the latest shift time in the observation information database 101, that is, the data of the latest section is acquired (hereinafter referred to as analysis data). ) And subsequent processing. Here, the latest needle changing speed is the newest time data among the previously estimated needle changing speeds. Therefore, the latest section data is time-series observation information (analysis data) from the latest needle shift time to the latest observation time.

  The change-of-needle shift detection process 402 is a process for detecting the presence or absence of a change-of-needle shift using statistical means from the analysis data, and since a number of patents have been filed (for example, Patent Document 2), description thereof is omitted here. To do. The result of the change-of-needle shift detection process is determined at 403, and when a change-of-needle shift is detected, the change-of-needle shift time is estimated by the process after 404.

  The idea is to perform a two-leg analysis from the latest observed data time to the latest needle changing speed time, using the needle changing speed time as an assumed value, and the estimated value with the best evaluation value is used as the estimated needle changing speed time. The details of the two-leg analysis will be described later, but using the time-series observation data of the two constant-velocity straight sections, the position coordinates of the target body at the reference time, the speed before and after the gear change, and the natural frequency are estimated. The sum of the normalized azimuth residual and the normalized frequency residual is output.

First, in process 404, an initial value Φ _init is set as the best evaluation value Φ. Since the evaluation values are a normalized azimuth residual and a normalized frequency residual, a smaller evaluation value is a better evaluation value. Therefore, Φ _init is set to a sufficiently large value. Next, in processing 405, the observation time t _new of the latest analysis data is set to the assumed value t _zig of the needle changing speed. In the two-leg analysis of the process 406, the target coordinate of the target body at the reference time, the speed before and after the needle shift, and the natural frequency are estimated with the needle change speed as the assumed value t _zig , and the evaluation value φ is output. Next, at the branch 407, the best evaluation value Φ is compared with the evaluation value φ calculated by the two-leg analysis. If the evaluation value φ calculated by the two-leg analysis is better, the processing 408 and the processing 409 are the best. The estimated value Φ and the estimated needle shift time T _zig that is the best evaluated value are updated. In process 410, the assumed value of the needle change speed change time is traced back by dt, and in step 411, if it is not past the needle change speed change oldest time _Tmax , process 406 and subsequent steps are repeated. Eventually, the needle change speed becomes T _zig and is stored in the needle change speed database 103 in processing 412.

The oldest time for changing the needle shift T _max is the oldest time in the latest section, and is the latest time for changing the needle changing speed. In order to repeat the process of FIG. 4 so as to go back in time, the assumed value of the needle change gear shift time is traced back by dt, and the process of FIG. 4 is the latest observation from the earliest time of the needle shift speed change T _max. It will be obvious that the time transition will be reversed if processing is performed toward the time. Moreover, dt should just be a value smaller than the resolution requested | required at the needle | hook change speed to estimate.

  FIG. 5 is a diagram for explaining the structure of the analysis history database 105. The analysis result is managed by a natural frequency and a motion vector of a constant speed straight section called a leg. The number of legs stored in the analysis history database is held at the number of stored legs. The leg has a structure for storing an analysis reference time, a target position X coordinate at the analysis reference time, a Y coordinate at the analysis reference time, a speed X component, and a speed Y component.

  FIG. 6 is a flowchart for explaining the processing of the target motion analysis unit 104. First, at branch 601, referring to the needle change gear shift database 103, if the number of needle change gear shifts is 0, a one-leg analysis of processing 602 is executed, and if the number of needle change gear shifts is one or more, a two-leg analysis of 603 is executed. .

  Here, 1-leg analysis applies a constant-velocity straight-ahead model as the target body motion model in the section from the latest needle changing time to the latest observation time, and the position of the target body at the analysis reference time. Coordinates, speed, and natural frequency are estimated. The two-leg analysis is a target body motion model, and travels straight at a constant speed in each of the section from the above-mentioned latest needle shift time to the estimated needle shift speed and the section from the estimated needle shift time to the latest observation time. The model is applied, and the position coordinates of the target body, the natural frequency, the course of each constant speed straight section, and the speed are estimated at the analysis reference time.

  Finally, the processing result 604 stores the analysis result in the analysis history database 105.

  FIG. 7 is a diagram illustrating the geometric relationship between the observation object and the target object, which explains the motion parameters estimated in the one-leg analysis. Reference numeral 701 represents the motion trajectory of the observation body, and 702 represents the motion trajectory of the target body. Reference numerals 703 to 705 indicate the position of the observation object that observed the sound wave emitted from the target object, and 705 represents the position of the observation object that acquired the latest observation data. The time when the latest observation data was acquired is used as the analysis reference time. . The parameters estimated in the one-leg analysis are the target body position X coordinate 707 at the analysis reference time, the target body position Y coordinate 708 at the analysis reference time, the speed X component 710, the speed Y component 711, and the target body natural frequency.

These parameters are estimated by minimizing Equation (1) by the least square method. Here, t represents the time normalized with the analysis reference time set to 0, C represents the speed of sound, and co and mo represent the course and speed of the observation body. Y _i and z _i are the observation direction and the observation frequency at time t _i , respectively, and equations (2) and (3) are the estimated direction and the estimated frequency, respectively. Xo _i and yo _i are the positions (xo _i, yo _i ) of the observation object at time t _i .

特に、式(2)(3)から構成される評価値(1)はパラメータx₁,x₂,x₃,x₄,x₅に対して非線形であるため、反復により解く必要があり、例えば、Levenberg-Marquardt法ではΔXが十分小さくなるまで式(7)(8)の計算繰り返し実施する。なお、λは定数、Ｉは単位行列である。またbは、ヤコビアン行列と残差ベクトルの積であり、ここでは、方位ヤコビアン行列の転置行列と方位残差ベクトルの積、および周波数ヤコビアン行列の転置行列と周波数残差ベクトルの積の二つの積の和である。

In particular, the evaluation value (1) composed of the equations (2) and (3) is nonlinear with respect to the parameters x ₁ , x ₂ , x ₃ , x ₄ , and x ₅ , and therefore needs to be solved by iteration, for example, In the Levenberg-Marquardt method, the calculations of equations (7) and (8) are repeated until ΔX becomes sufficiently small. Λ is a constant, and I is a unit matrix. B is the product of the Jacobian matrix and the residual vector. Here, the product of the transposed matrix of the azimuth Jacobian matrix and the product of the azimuth residual vector, and the product of the transposed matrix of the frequency Jacobian matrix and the product of the frequency residual vector. Is the sum of

式(8)のA、Bは方位情報及び周波数情報のヤコビアン行列であり、次式となる。

A and B in Equation (8) are Jacobian matrices of azimuth information and frequency information, and are given by the following equations.

図８はｎレグ解析において等速直進区間が２区間の場合に、推定する運動パラメータを説明する観測体と目標体の幾何学的関係を示した図である。801は観測体の運動軌跡、802は目標体の運動軌跡を表している。803は最古の観測体の位置座標、804は推定変針変速時刻における観測体の位置座標、805は最新の観測体の位置座標を表している。また806は、推定した変針変速時刻における目標体の位置であり、そのX座標を807で、そのY座標を808で表している。

FIG. 8 is a diagram showing the geometric relationship between the observation object and the target object for explaining the motion parameters to be estimated when there are two constant velocity straight sections in the n-leg analysis. Reference numeral 801 represents the motion trajectory of the observation body, and 802 represents the motion trajectory of the target body. Reference numeral 803 denotes the position coordinates of the oldest observation object, reference numeral 804 denotes the position coordinates of the observation object at the estimated time of changing the needle shift, and reference numeral 805 denotes the latest position coordinates of the observation object. Reference numeral 806 denotes the position of the target body at the estimated time of changing the needle shift, and its X coordinate is represented by 807 and its Y coordinate is represented by 808.

  The analysis reference time is the latest observation time, and the parameters estimated in the two-leg analysis are the X coordinate 813 of the target body position 812 at the analysis reference time, the Y coordinate 814 of the target body position 812 at the analysis reference time, and the speed 809 before changing the needle shift. X component 810, Y component 811 of speed 809 before changing gear speed, X component 816 of speed 815 after changing speed, Y component 817 of speed 815 after changing speed, and natural frequency.

These parameters are estimated by minimizing equation (1) as in the case of 1-leg analysis. In the flowchart of FIG. 4, S (X) in Expression (1) in the two-leg analysis is represented by Φ. However, the estimated azimuth and the estimated frequency are expressed by equations (21) to (24), where the estimated needle changing time is t _z (indicated by T _{zig in} the flowchart of FIG. 4).

Levenberg-Marquardt法等により計算するためのヤコビアンは式(30)〜(56)となる。

Jacobians for calculation by the Levenberg-Marquardt method and the like are expressed by equations (30) to (56).

  As an embodiment, a configuration example of the target motion analysis apparatus according to the present invention is shown in FIG. As shown in FIG. 9, the target motion analysis device 901 can be realized by an electronic computer comprising a CPU 904, a memory 905, an output device 906, I / F (interfaces) 907 and 908, and an input device 909. Connected to sensor 902 and navigation sensor 903. It may be configured as a dedicated device due to restrictions on the size and calculation speed of the target motion analysis device.

  The acoustic sensor 902 receives a sound wave radiated from the target body, and inputs the arrival direction and frequency of the sound wave to the target motion analysis apparatus 901. The navigation sensor 903 inputs the position, speed, and course of the observation body to the target motion analysis device 901 using an electromagnetic log, gyroscope, GPS, or the like. The arrival direction, frequency, position of the observation object, speed, and course of the input sound wave are stored in the observation information database in the memory 905. The change-of-needle shift detection unit and the target motion analysis unit perform processing based on the above-described method, and store the results in the change-of-change gear database and the analysis history database.

  The contents of the observation information database, the needle change speed change database, and the analysis history database held on the memory 905 are displayed on an output device 906 such as a CRT and presented to the operator. The operator refers to the contents of the observation information database, the needle change gear shift database, and the analysis history database output to the output device 906. It can also be stored.

  For example, as shown in FIG. 10, a graph with time on the horizontal axis and observation direction on the vertical axis is displayed on the output device 906. FIG. 10 shows the arrival direction of the sound wave observed by the acoustic sensor when the target movement pattern is as shown in Table 1, and it can be seen from FIG. 10 that the tendency of the observation direction changes in 700 seconds.

しかしながら、実際に観測される方位は音波伝搬中の屈折の影響や音響センサにおける計測誤差もあり、図１０のように鮮明な方位の傾向変化を読み取る事は難しいため、本装置は統計的な手段により変針変速を看破する変針変速検出部を備えている。

However, since the azimuth actually observed is affected by refraction during sound wave propagation and measurement errors in the acoustic sensor, it is difficult to read a clear change in azimuth as shown in FIG. Thus, a change-of-needle shift detection unit for detecting change-of-change gears is provided.

  FIG. 13 is a simulation result of the target motion analysis method according to the present embodiment, in which 1301 is the track of the observation body, 1302 is the track of the target body, and 1303 is the target body estimation when the target motion analysis is performed with an analysis period of 30 seconds. Position coordinate history, 1304 is the relative distance between the observation object and the target object, 1305 is the estimated relative distance between the observation object and the target object based on the target motion analysis result, 1306 is the speed of the target object, 1307 is the target object estimation based on the result of the target motion analysis Speed, 1308 represents the course of the target body, and 1309 represents the estimated course of the target body based on the result of the target motion analysis. In 900 seconds after the simulation starts, the target changes its course from 180 ° to 150 °, and the analysis results are output stably after the target has changed. The conditions of this simulation are the same as those in FIGS. 11 and 12, and it can be seen that stable analysis can be performed even when the target body is changing speed by the present embodiment.

It is a block diagram for performing target motion analysis. It is a figure explaining the structure of an observation information database part. It is a figure explaining the structure of a needle change speed change database part. It is a flowchart figure of a needle change speed change detection part. It is a figure explaining the structure of an analysis history database part. It is a flowchart figure of a target exercise | movement analysis part. It is the figure which showed the geometric relationship of the observation body and target body in the target motion analysis of a constant-velocity linear model. It is the figure which showed the geometric relationship of the observation body and target body in the target motion analysis of a variable speed transmission model. It is an example of composition of a target motion analysis device. It is the graph which showed the arrival azimuth | direction of the sound wave when a target changes a needle in time series. It is a simulation result at the time of applying the target motion analysis by the constant-velocity linear model which is a prior art with respect to the target body which changes a needle | hook. It is a simulation result at the time of applying a target motion analysis to the target body which changes a needle | hook using the observation data after the change-of-speed-shift detection which is a prior art. It is a simulation result at the time of applying a target motion analysis to the target object to change a course.

Explanation of symbols

  101 ... Observation information database unit, 102 ... Change gear shift detection unit, 103 ... Change gear shift database unit, 104 ... Target motion analysis unit, 105 ... Analysis history database unit, 901 ... Target motion analysis device, 902 ... Acoustic sensor, 903 ... Navigation Sensor, 904 ... CPU, 905 ... Memory, 906 ... Output device, 907-908 ... Interface, 909 ... Input device.

Claims (4)

The arrival direction and frequency of the sound wave radiated from the target object are observed in time series by an acoustic sensor, and the position of the target object at an arbitrary reference time using the accumulated time series observation direction data and time series observation frequency data In a target motion analysis method for estimating a motion vector including a course and speed,
(a) In response to the detection of the target changing gear shift, when the changing gear shift is detected, assuming the changing needle shift time,
(b) Assuming that each of the sections before and after the assumed needle changing speed is regarded as a two-leg motion in which the target body is moving straight at a constant speed, it is the latest data among the previously estimated needle changing speeds. The constant-velocity linear model is applied to each of the section from the assumed needle change gear shift time to the assumed change gear shift time and the section from the assumed needle change gear shift time to the latest observation time by a nonlinear least square method. Two-leg analysis for estimating the motion parameter is performed to obtain time-series estimated azimuth data and time-series estimated frequency data before and after the assumed gear change time ,
(c) Evaluating the residual between the time-series observation azimuth data and the time-series estimated azimuth data before and after the assumed change-in-shift time and the sum of the residuals between the time-series observation frequency data and the time-series estimated frequency data A target motion analysis method that estimates, as values, the position of the target body at the arbitrary reference time, the course and speed before and after the shift change by an optimization method using a derivative. Wherein (a) veering transmission time assumptions, the while changing in the past direction from the detection time of veering speed, repeat the steps (b) and (c) until the evaluation is worse method, wherein (a assumed veering desired motion analysis method according to claim 1, wherein the evaluation value is estimated as the veering shift time the smallest time of transmission time in). An acoustic sensor for observing the arrival direction and frequency of sound waves radiated from the target in time series, and the arrival direction and frequency of the sound waves observed by the acoustic sensor as time series observation direction data and time series observation frequency data A database to store, and
(a) When detecting the change of needle shift in response to detection of the change of needle shift of the target body, the process of assuming the time of the change of needle shift, and (b) each of the sections before and after the assumed change of gear change speed. wherein the target body is regarded as 2 leg exercise that constant velocity rectilinear motion, and the section from the most recent veering shift time to the assumed veering shift time the newest data among the veering shift time previously estimated, the A two-leg analysis is performed to estimate a motion parameter by a non-linear least-squares method by applying a constant-velocity straight-ahead model to each of the sections from the assumed needle changing time to the latest observation time, and for each section, the time-series observation azimuth data processing for determining the sequence estimation frequency data when based on the sequence estimation orientation data the time-series observation frequency data when based on, and (c) the assumed veering shift time the time system of the front and rear The position of the target body at an arbitrary reference time, using as an evaluation value the residual between the sequence observation azimuth data and the time-series estimated azimuth data and the sum of the residuals between the time-series observation frequency data and the time-series estimated frequency data A target motion analysis apparatus having a processor for executing a process of estimating the course and speed before and after the change-of-change gear speed by an optimization method using a derivative. The processor repeats the processes of (b) and (c) until the evaluation becomes a deteriorating method while changing the assumed time of change-of-needle shift from the detection time of change-of-change to the past direction, and as a result of the evaluation, The target motion analysis apparatus according to claim 3, wherein the target gear shifting time at which the value is minimized is output.

Images