CN110376574B - A Target Tracking Method Based on Multi-base Passive Sonar Observation Data - Google Patents

Abstract

The invention relates to a target tracking method based on multi-base passive sonar observation data. Based on the measurement data of passive multi-sensors, the tracker quickly converges on the estimation of the current state of the target, and the convergence effect is very stable. It is a method with good concealment , a powerful and robust tracker.

Description

A Target Tracking Method Based on Multi-base Passive Sonar Observation Data

technical field

The invention belongs to the field of sensor information fusion, in particular to a target tracking algorithm based on multi-base passive sonar observation data.

Background technique

Passive sonar detection locates the target by receiving the noise radiated by the target in the water. Because sonar itself does not emit signals, it has the advantages of good concealment and not easy to be attacked, and has attracted much attention. A single passive sensor can only observe the orientation of the target, but cannot obtain the distance information, which belongs to incomplete observation. To complete the target tracking task, the information of multiple sensors needs to be fused, and the target tracker based on passive measurement has not always had a relatively complete process. Therefore, the research on passive multi-sensor trackers is very necessary.

Contents of the invention

The technical problem solved by the present invention is: in order to make up for the deficiencies of the prior art, the present invention designs a target tracking method based on multi-base passive sonar observation data, re-describing the motion model and observation model in the target tracker, and obtains a A reliable target tracking method suitable for receiving passive measurements.

The technical scheme of the present invention is: a kind of target tracking method based on multi-base passive sonar observation data, comprises the following steps:

Step 1: Define the application scenario: several sonar base stations for passive observation are distributed in space, and the nodes of each base station form a network through self-organization. In the area between several sonar base stations, there are multiple targets. The target radiates acoustic signals to the surroundings indiscriminately, and the passively observed sonar base station receives the acoustic signals, and uses the output results of signal processing (filtering, beamforming, matched filtering) to obtain the relative orientation and receiving frequency of the target;

Step 2: According to the application scenario of step 1, create a target motion model and an observation model, wherein the target motion model is an approximate constant velocity model, and the observation model is a nonlinear system;

Step 3: According to the motion model and observation model in step 2, design a nonlinear filter. The design is as follows: Apply the extended Kalman filter method to convert the nonlinear system into an approximate linear model around the filter value. When the filter performs measurement data association, there are two time series structures: the association of measurement data returning multiple directions at the same time, and the association of continuously returning measurement data in the uniform forward time series can be attributed to the same type of problem. Solve, and finally get a complete target tracking process, which can track the position coordinates of the target.

A further technical solution of the present invention is: the target motion model is an approximate uniform velocity model.

Invention effect

The technical effect of the present invention is that: based on the measurement data of passive multi-sensors, the tracker can quickly converge on the estimation of the current state of the target, and the convergence effect is very stable. It is a tracker with good concealment, powerful functions and stable performance.

Description of drawings

Figure 1 is a schematic diagram of passive observation of targets by multiple sensors;

Figure 2 is an effect diagram of the present invention.

detailed description

The process of estimating the current state parameters of the tracked target and processing the measurement data received by the sensor is called the target tracking process.

See Figure 1-Figure 2, a scheme that applies passive sensor data to the target tracking algorithm framework, including the creation of motion models and observation models. Since the observation model is a nonlinear system that needs to be transformed into a linear system, different time sequences under multi-sensor conditions Get measurement data processing. It is characterized in that: according to the unique application scenarios of the present invention, corresponding observation and motion models are created; nonlinear filters are redesigned based on the nonlinear observation models; a timing structure different from traditional filters is adopted for multi-sensor application scenarios.

The state of the target includes the position, velocity and radiation frequency of the target; the measurement of the target includes the relative orientation of the target and the receiving frequency; the Doppler effect is considered in the receiving frequency. Since the measurement system of passive tracking is nonlinear, the linear Kalman filter used in the traditional target tracking process is not suitable, so it is replaced by a classical nonlinear filter - the extended Kalman filter.

In the case of passive multi-sensor tracker observation, it is necessary to correlate the measurement data received by the sensors, subdividing into two data correlation tasks, one is the correlation of multiple orientation measurements returned at the same time, and the other is The first is the association of time series measurements. The time difference Δt between frames in the filter is a variable controlled by the data source. For data from different sensors at the same time, Δt can be 0, so the two associations Tasks can actually be classified as one.

Main content of the present invention has:

1. Application scenarios of the present invention:

The scene of multi-sonar (sensor) passive observation described in the present invention is shown in Fig. 1, and the sonar base stations of a plurality of fixed passive observations are distributed in space, and the target radiates acoustic signals to the surroundings indiscriminately. The base station receives the signal and uses the signal processing algorithm to obtain some data (measurement) related to the state of the target, including the relative orientation and receiving frequency of the target. These data can be used as the basis for the tracker to discover and track the target.

2. The target motion state has the characteristics of unpredictable and random changes, and the target motion model of the current tracking system may not match the actual motion of the target. These factors will increase the difficulty of target state estimation and prediction, so establish a suitable target motion Models are very important. A commonly used mathematical model for moving targets is given here: the approximate constant velocity (NCV) model.

Step 1: The motion model can be described by the state equation as:

X _k+1 ＝Φ _k X _k +ω _k

频率信息

X_k为k时刻的系统状态向量，X_k+1为(k+1)时刻的系统状态向量，Φ_k为k时刻的状态转移矩阵，ω_k为k时刻的零均值，方差为Q_k的高斯过程噪声，将目标运动速度建模为一个维纳过程，则x轴，y轴上的速度不确定参数分别为q_x,q_y，频率不确定参数为q_f，得到协方差矩阵Q_k的具体表示。In the formula, X is the system state vector, including position coordinates (x _k , y _k ), x-axis, y-axis velocity components

frequency information

X _k is the system state vector at k time, X _k+1 is the system state vector at (k+1) time, Φ _k is the state transition matrix at k time, ω _k is the zero mean at k time, and the variance is Q _k Gaussian process noise, model the speed of the target as a Wiener process, then the speed uncertainty parameters on the x-axis and y-axis are q _x , q _y respectively, and the frequency uncertainty parameters are q _f , and the covariance matrix Q _k is obtained specific representation.

Step 2: Observation equation:

v_k表示观测噪声，c表示声速。

表示观测站的坐标位置，r_k表示观测站与目标的距离，d_k/r_k表示目标关于观测站的径向速度。Passive multi-sensors return the target’s azimuth and frequency, assuming that Z _k is the system observation vector at time k, the azimuth observation θ _k of the target at k time and the frequency observation f _k are independent of each other, and both have Gaussian errors, respectively

v _k represents the observation noise, and c represents the speed of sound.

Indicates the coordinate position of the observation station, rk represents the distance between the observation station and the target, and d _k / _{r k represents} the radial velocity of the target relative to the observation station.

The atan2 function in the formula represents the two-parameter arctangent function, and the value range is (-π, π].

Step 3: For the active tracker, the first measurement can be directly converted into position coordinates, and the covariance of this coordinate is used to construct the matrix P(1|1). For the passive measurement, the first measurement It is the azimuth and frequency information, not the direct position coordinate information. Since the position coordinate information is vague, the estimation error is relatively large, so the initial value of P(1|1) needs to be set very large.

The initial state is described as follows:

P(1|1)表示估计误差的协方差矩阵，其中

是位置坐标的高斯误差，

是速度的高斯误差，

是频率的误差。Among them, X(1|1) is the predicted state vector of the system at the initial moment, assuming that the position coordinates are (x ₁ , y ₁ ), the speed is 0, and the frequency is

P(1|1) represents the covariance matrix of the estimation error, where

is the Gaussian error of the position coordinates,

is the Gaussian error of the velocity,

is the frequency error.

Step 4: Nonlinear Filtering

In step 2, it is known that the orientation and frequency are obtained from the sensor, and the H matrix in the observation equation is nonlinear, so a nonlinear filter—the most classic extended Kalman filter is used to approximate the nonlinear system to a linear system, and re- Describe the system prediction and update process, and the derivation is as follows:

X(k+1|k) is the system prediction state vector at time k+1, and P(k+1|k) is the covariance matrix representing the prediction error.

The linearization of the H matrix in the observation equation is as follows:

in

in

J(k+1) represents the equivalent observation matrix of the k+1th time series, which is the partial derivative of the observation function H.

P(k+1|k+1)=(I-L(k+1)J(k+1))P(k+1|k)

X(k+1|k+1) is the state update vector at time k+1, and P(k+1|k+1) represents the covariance matrix corresponding to the state vector update step.

A problem with the tracker involved in 2 is that the tracker cannot converge if the measurements are always from the same fixed sonar. This problem can be avoided when there are multiple fixed sonar base stations. The method is to dynamically adjust the time stamp of the filter. When the measured time changes, the delay Δt will be changed. Otherwise, for the same time obtained by different base stations The sensor data Δt=0, the advantage of doing this is that the filter of passive positioning is completed together, the passive positioning method of multiple targets under passive observation is difficult to design, this can avoid this process. (This section is the limiting condition of Δt in the filter)

Implementation example: This example mainly demonstrates the working effect of the filter under non-uniform timing. ,1000)m and move to the upper right. The filter gets the bearing of the target from the sensor. The sensor estimates the target orientation error σ _θ =0.1°, q _x =q _y =300. It can be seen from Figure 2 that the solid line represents the real trajectory of the target, and the dotted line represents the filtered tracking result. For any set filter initial value (1500, 1500) m away from the target, it can be seen that the solid line and the dotted line overlap quickly. The filter can converge quickly, and in the subsequent tracking process, the two lines can always overlap well, that is, the error between the tracking result and the real track is small, and the tracker can track the target well.

Claims (1)

1. A target tracking method based on multi-base passive sonar observation data is characterized by comprising the following steps:

the method comprises the following steps: defining an application scene: a plurality of passive observation sonar base stations are distributed in a distributed manner in space, nodes of each base station form a network in a self-organizing manner, a plurality of targets are arranged in an area among the plurality of sonar base stations, the targets radiate acoustic signals to the surroundings without distinction, the passive observation sonar base stations receive the acoustic signals and obtain the relative direction and the receiving frequency of the targets by using the output result of signal processing, and the method for signal processing comprises filtering, beam forming and matched filtering;

step two: according to the application scene in the first step, a target motion model and an observation model are established, wherein the target motion model is an approximate uniform velocity model, and the observation model is a nonlinear system;

step three: designing a nonlinear filter according to the motion model and the observation model in the step two, wherein the design is as follows: applying an extended Kalman filtering method to convert the nonlinear system into an approximate linearized model around the filtered value; when the filter performs measurement data correlation, there are two timing structures: the measurement data of a plurality of directions are returned at the same time for correlation, and the correlation of the measurement data continuously returned in uniform forward time sequence is solved by solving the same problem, finally a complete target tracking process is obtained, and the position coordinate of the target can be tracked;

step four: when the measured data correlation is performed in the filter in the third step, two timing structures exist: the measurement data of a plurality of directions are returned at the same time for correlation, the measurement data which are continuously returned in uniform forward time sequence are correlated, a correlation frame with universality is designed to complete two different correlation tasks by setting the time difference delta t between frames according to different time sequence structures, and finally a complete target tracking process is obtained and the position coordinates of a target can be tracked.

Images