Doppler Radar Sequential Smoothing Variable Structure Filtering Method and Device

Abstract

The present application provides a sequential smooth variable structure filtering method and device for a Doppler radar, belonging to the technical field of radar data processing, which can simultaneously solve the problems of the nonlinear underdetermined observation model of the Doppler radar and the robust tracking problem under the condition of uncertain target motion models. The method adopts the strategies of decoupling processing and sequential estimation of the target position measurement and radial velocity measurement of the Doppler radar to solve the nonlinear underdetermined observation model problem: in the first measurement conversion module and the first state estimator, the first-level target state estimation is obtained by using the nonlinear target position measurement and the generalized smooth variable structure filtering method; in the second-level measurement conversion module and the second state estimator, a pseudo-measurement is constructed by using the radial velocity measurement to update the first-level target state estimation, and the final posterior state estimation of the target in the current frame is obtained. Among them, the generalized smooth variable structure filtering method ensures the robustness under the condition of model uncertainty. Therefore, the present application can improve the target tracking accuracy of the Doppler radar and maintain the robust estimation performance under the condition of uncertain target motion models at the same time.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of radar data processing, and more specifically, to a sequential smooth variable structure filtering method and device for a Doppler radar. Background Art

[0002] A radar target tracking device is a device that uses information such as target distance, azimuth angle, Doppler frequency offset, etc. contained in radar echoes to continuously and effectively estimate the number of targets and the motion states such as position, speed, and acceleration of the targets. It is widely used in fields such as autonomous driving, air traffic control, meteorological monitoring, security, and national defense. Among them, a Doppler radar can observe the radial velocity information of target motion, and thus increase the effective information dimension in the radar target state estimation link, greatly improving the target tracking performance.

[0003] The target tracking algorithm or device of a Doppler radar needs to establish a radar observation model and a target state transition model. However, since the observation model of a Doppler radar is a highly non-linear observation model, and a highly non-linear model will cause filters commonly used in conventional radar tracking methods, such as a Kalman filter (KF), etc., to generate large non-linear estimation errors, and even lead to filter divergence. At the same time, the target state transition model of a Doppler radar has uncertainty, and traditional Bayesian filtering methods, including a Kalman filter (KF), etc., all rely on an accurate description of the target motion model. If the deviation between the target state transition model and the real target motion model is large, for example, the target undergoes a strong maneuvering motion, it will cause the tracking error to increase sharply and even cause the filter to diverge and lose the target.

[0004] To solve the errors caused by the non-linear observation model of a Doppler radar, a series of non-linear filtering methods are also applied when a Doppler radar is used, including extended Kalman filtering (EKF), unscented Kalman filtering (UKF), and particle filtering (PF), and also including a sequential Kalman filtering (SKF) method with high estimation accuracy and low computational complexity. However, these traditional Bayesian methods, including a sequential Kalman filter and its numerous improved forms, although they can reduce non-linear errors, cannot solve the uncertainty problem of the target state transition model.

[0005] To solve the uncertainty problem of the target state transition model of Doppler radar, methods such as robust Kalman filtering, H-infinite filtering, and the recently popularized smooth variable structure filtering (SVSF) are often used. SVSF is a robust filtering method under model uncertainty conditions, which can ensure the boundedness of the state estimation error and has a relatively low computational complexity. However, the standard SVSF requires a linear observation model and a full-rank observation matrix to achieve a bijective mapping from the target state space to the observation space. Since the observation model of Doppler radar is non-linear, in order to solve the non-linear observation model problem, attempts have been made to substitute the Jacobian matrix for the non-linear function into the standard SVSF structure, but this will introduce a new problem of inverting an underdetermined matrix, resulting in a large estimation error of the target velocity. Therefore, the SVSF method still cannot be applied to Doppler radar devices.

[0006] Therefore, the accuracy problem caused by the non-linear underdetermined observation model of Doppler radar and the robust tracking problem under the condition of uncertain target motion models still need to be urgently solved. Summary of the Invention

[0007] The embodiments of the present application provide a sequential smooth variable structure filtering method, device, equipment, and storage medium for Doppler radar, aiming to improve the tracking accuracy of the observed target state of Doppler radar and maintain robust estimation performance under the condition of uncertain target motion models.

[0008] In a first aspect, the embodiments of the present application provide a sequential smooth variable structure filtering method for Doppler radar, and the method includes the following steps:

[0009] A first measurement conversion module converts the non-linear target position measurement of the current frame Doppler radar into a linear position measurement in the Cartesian coordinate system;

[0010] A first state estimator uses a generalized smooth variable structure filter and the linear position measurement in the Cartesian coordinate system to obtain a first-level target state estimate;

[0011] A second measurement conversion module constructs a target range-radial velocity product pseudo-measurement using the target radial velocity measurement of the current frame Doppler radar;

[0012] A second state estimator updates the first-level target state estimate based on the target range-radial velocity product pseudo-measurement to obtain the final posterior state estimate of the current frame target.

[0013] The first measurement conversion module uses an unbiased measurement conversion method to convert the non - linear target position measurement vector in the three - dimensional spherical coordinates or two - dimensional polar coordinates of the Doppler radar into the linear position measurement vector in the Cartesian coordinate system, and calculates the first conversion deviation and the first conversion covariance.

[0014] The first state estimator uses a generalized smooth variable - structure filter and the linear position measurement in the Cartesian coordinate system to obtain the first - level target state estimation, including the following steps:

[0015] Calculate the prior estimate of the target state, the prior estimate of the radar measurement, and the prior state - estimation covariance matrix for the current frame;

[0016] According to the linear position measurement of the current frame, the first conversion deviation, the first conversion covariance, and the prior estimate of the radar measurement, calculate the prior position measurement error;

[0017] According to the prior position measurement error, calculate the first innovation gain term;

[0018] According to the prior estimate of the target state, the prior state - estimation covariance matrix, the prior position measurement error, and the first innovation gain term, calculate the posterior estimate of the target state and the posterior state - estimation covariance matrix for the current frame, and use the posterior estimate of the target state and the posterior state - estimation covariance matrix for the current frame as the first - level target state estimation.

[0019] The second measurement conversion module uses the target distance measurement and the target radial velocity measurement of the Doppler radar to construct the range - radial velocity product pseudo - measurement, and uses an unbiased conversion method to calculate the second conversion deviation, the second conversion variance, and the second conversion covariance of the range - radial velocity product pseudo - measurement.

[0020] The second state estimator updates the first - level target state estimation based on the target range - radial velocity product pseudo - measurement to obtain the final posterior state estimation of the target for the current frame, including the following steps:

[0021] Perform pre - whitening processing on the range - radial velocity product pseudo - measurement to obtain the pre - whitened pseudo - measurement, the deviation of the pre - whitened pseudo - measurement, and the variance of the pre - whitened pseudo - measurement;

[0022] Take the first - level target state estimation, the pre - whitened pseudo - measurement, the deviation of the pre - whitened pseudo - measurement, and the variance of the pre - whitened pseudo - measurement as inputs, construct a locally approximated linear minimum mean - square error estimator, and update the first - level target state estimation to obtain the final posterior state estimation of the target for the current frame.

[0023] Second aspect, an embodiment of the present application provides a Doppler radar sequential smoothing variable structure filtering device, where the filtering device includes a first measurement conversion module, a first state estimator, a second measurement conversion module, and a second state estimator, where:

[0024] The first measurement conversion module is configured to convert the non-linear target position measurement of the current frame Doppler radar into a linear position measurement in the Cartesian coordinate system;

[0025] The first state estimator is configured to obtain a first-level target state estimate by using a generalized smoothing variable structure filter and the linear position measurement in the Cartesian coordinate system;

[0026] The second measurement conversion module is configured to construct a target range-radial velocity product pseudo-measurement by using the target radial velocity measurement of the current frame Doppler radar;

[0027] The second state estimator is configured to update the first-level target state estimate based on the target range-radial velocity product pseudo-measurement to obtain the final posterior state estimate of the target in the current frame.

[0028] The first measurement conversion module is configured to convert the non-linear target position measurement vector in the three-dimensional spherical coordinates or two-dimensional polar coordinates of the Doppler radar into the linear position measurement vector in the Cartesian coordinate system by using an unbiased measurement conversion method, and calculate a first conversion deviation and a first conversion covariance.

[0029] The first state estimator includes:

[0030] A first calculation unit configured to calculate a prior estimate of the target state, a prior estimate of the radar measurement, and a prior state estimate covariance matrix of the current frame;

[0031] A second calculation unit configured to calculate a prior position measurement error according to the linear position measurement of the current frame, the first conversion deviation, the first conversion covariance, and the prior estimate of the radar measurement;

[0032] A third calculation unit configured to calculate a first innovation gain term according to the prior position measurement error;

[0033] A fourth calculation unit configured to calculate a posterior estimate of the target state and a posterior state estimate covariance matrix of the current frame according to the prior estimate of the target state, the prior state estimate covariance matrix, the prior position measurement error, and the first innovation gain term, and use the posterior estimate of the target state and the posterior state estimate covariance matrix of the current frame as the first-level target state estimate.

[0034] The second measurement conversion module uses the target distance measurement and target radial velocity measurement of the Doppler radar to construct the distance-radial velocity product pseudo-measurement, and uses the unbiased conversion method to calculate the second conversion deviation, second conversion variance, and second conversion covariance of the distance-radial velocity product pseudo-measurement.

[0035] The second state estimator includes:

[0036] A pre-whitening processing unit for pre-whitening the distance-radial velocity product pseudo-measurement to obtain a pre-whitened pseudo-measurement, the deviation of the pre-whitened pseudo-measurement, and the variance of the pre-whitened pseudo-measurement;

[0037] A posterior state estimation unit for using the first-level target state estimation, the pre-whitened pseudo-measurement, the deviation of the pre-whitened pseudo-measurement, and the variance of the pre-whitened pseudo-measurement as inputs to construct a locally approximated linear minimum mean square error estimator to update the first-level target state estimation and obtain the final posterior state estimation of the target in the current frame.

[0038] Advantageous effects:

[0039] The sequential smoothing variable structure filtering method proposed by the present invention has two advantages:

[0040] First, this method has the robustness advantage under model uncertainty. Based on the sliding mode variable structure control theory, this method can ensure the boundedness of the tracking error under the condition that the target motion model is unknown, and solves the problem that the tracking filtering error of traditional non-linear Bayesian filters (such as Kalman filters and sequential Kalman filters, etc.) increases sharply or even the filtering diverges when there are modeling errors in the target state transition model.

[0041] Specifically, first, the non-linear target position measurement of the Doppler radar is converted into a linear position measurement in the Cartesian coordinate system, and then a linear generalized smoothing variable structure filter can be constructed to achieve robust estimation of the target state. By further updating and correcting the output first-level target state estimation, the robust state estimation performance under model uncertainty is ensured.

[0042] Second, this method uses the state space sequential estimation method to divide the observation vector of the Doppler radar into two parts, including the target position measurement and the target radial velocity measurement, and sequentially updates the state estimation with these two parts of measurements, solving the underdetermined problem of the non-linear model and non-full rank observation matrix of the existing smoothing variable structure filtering method for the Doppler radar. Furthermore, the radial velocity measurement information of the Doppler radar can be fully utilized to effectively improve the target state estimation accuracy of the smoothing variable structure filter.

[0043] Specifically, the sequential smooth variable structure filter solves the problem that the standard smooth variable structure filter cannot utilize the non-linear Doppler measurement, and also solves the underdetermined problem of inverting the non-full rank observation matrix when the improved smooth variable structure filter in the existing literature uses the Jacobian matrix for local linearization operation. Therefore, the method of the present invention makes more effective use of the Doppler velocity information compared with the existing smooth variable structure filtering methods, thereby significantly improving the accuracy of target state tracking.

[0044] In summary, the method of the present invention can significantly improve the accuracy of Doppler radar target state estimation, while ensuring the filtering robustness under the condition of model uncertainty, and has good application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0046] Figure 1 is a flowchart of the steps of the filtering method proposed in an embodiment of the present application;

[0047] Figure 2 is a target maneuvering trajectory diagram in the simulation experiment proposed in an embodiment of the present application;

[0048] Figure 3 is a schematic diagram of the change of the coordinate error of the simulation results with the tracking time proposed in an embodiment of the present application;

[0049] Figure 4 is a schematic diagram of the change of the velocity error of the simulation results with the tracking time proposed in an embodiment of the present application;

[0050] Figure 5 is a functional module diagram of the filtering device proposed in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present application belong to the scope of protection of the present application.

[0052] Refer to Figure 1, showing the step flowchart of a Doppler radar sequential smoothing variable structure filtering method in an embodiment of the present invention. This method is applied to a Doppler radar sequential smoothing variable structure filtering device, and the filtering device includes a first measurement conversion module, a first state estimator, a second measurement conversion module, and a second state estimator. The method may specifically include the following steps:

[0053] S101: The first measurement conversion module converts the non-linear target position measurement of the current frame Doppler radar into a linear position measurement in the Cartesian coordinate system.

[0054] Specifically, let the scanning period of the Doppler radar be T, and the measurement data vector of the current frame k of the Doppler radar for the target state be z m (k), and the measurement data includes the distance r m (k), the azimuth angle θ m (k), the elevation angle and the radial velocity

[0055] According to the structure of the measurement data of the Doppler radar, the measurement data vector of the current frame k for the target state, z m (k) is divided into the target position measurement and the target radial velocity measurement, and the expression form is as follows:

[0056]

[0057] In the formula, is the target position measurement including the distance, azimuth angle, and elevation angle, is the target radial velocity measurement, k represents the frame number of the Doppler radar, k = 1, 2, 3...

[0058] According to the type of the Doppler radar, the error standard deviation of each measurement variable can be viewed. Among them, σ r represents the observation error standard deviation of the distance, represents the observation error standard deviation of the radial velocity, represents the observation error standard deviation of the elevation angle, σ θ represents the observation error standard deviation of the azimuth angle, and ρ is the noise correlation coefficient between the distance measurement and the radial velocity measurement.

[0059] The target position measurement obtained by the Doppler radar is non-linear, and the first measurement conversion module needs to convert the non-linear target position measurement into a linear position measurement in the Cartesian coordinate system.

[0060] Specifically, the unbiased measurement conversion method (UCM) is adopted to convert the non-linear target position measurement in the spherical coordinate system or polar coordinate system of the Doppler radar into a linear position measurement in the Cartesian coordinate system The first conversion deviation μ p = [μ x , μ y , μ z T and the first covariance R p (k);

[0061] where r m represents the distance, θ m represents the azimuth angle, represents the elevation angle, and x, y, and z are the three coordinate axes in the Cartesian coordinate system.

[0062] It should be noted that in this embodiment, a three-dimensional Doppler radar is taken as an example. In other embodiments, for example, for a two-dimensional scenario, the effective dimensions can be directly intercepted.

[0063] S102: The first state estimator uses a generalized smooth variable structure filter and the linear position measurement in the Cartesian coordinate system to obtain a first-level target state estimate.

[0064] The smooth variable structure filter is a robust filtering method under model uncertainty conditions, which can ensure the boundedness of the state estimation error and has a relatively low computational complexity. However, the standard smooth variable structure filtering method requires a linear observation model and a full-rank observation matrix to achieve a bijective mapping from the target state space to the observation space, which cannot be used for the non-linear measurement of the Doppler radar. This method converts the non-linear target position measurement into a linear position measurement, so that the generalized smooth variable structure filter can be used for data processing, thus ensuring the robust state estimation performance under model uncertainty conditions.

[0065] The first state estimator obtains a first-level target state estimate, which specifically includes the following steps:

[0066] S1021: Calculate the prior estimate of the target state, the prior estimate of the radar measurement, and the prior state estimation covariance matrix for the current frame k.

[0067] Specifically, the prior estimate of the target state The prior estimate of the radar measurement and the prior state estimation covariance matrix P p (k|k - 1) are determined according to the following formula:

[0068]

[0069]

[0070]

[0071] In the formula, ​is the state estimation variable, is the preset state transition matrix, is the preset control input matrix, and u is the known control input variable, is the observation matrix, m and n respectively represent the dimensions of the state variable and the observation variable, and Q and Γ are the process noise covariance and its coefficient matrix, and are different from the target state transition models F and G that require accurate knowledge in traditional Bayesian filters, and allow for a norm-bounded model error compared to the true value.

[0072] Since the decorrelated unbiased measurement conversion method is applied in this method to convert the non-linear target position measurement into a linear position measurement, the observation matrix is linear time-invariant and is superior to the traditional smooth variable structure filter that directly uses non-linear position measurements.

[0073] S1022: Calculate the prior position measurement error based on the linear position measurement of the current frame, the first conversion deviation, the first conversion covariance, and the radar measurement prior estimate.

[0074] Specifically, the prior position measurement error e p (k|k - 1) is determined according to the following formula:

[0075]

[0076] where, is the linear position measurement, and μ p ( K ) is the first conversion deviation; is the radar measurement prior estimate.

[0077] S1023: Calculate the first innovation gain term according to the prior position measurement error.

[0078] Specifically, the first innovation gain term can be determined by the following method:

[0079] First, calculate the mixed error term:

[0080]

[0081]

[0082] where, e z (k - 1|k - 1) is the posterior measurement error of the (k - 1)-th frame, and |·| ABS represents taking the absolute value of each element of the vector, and is a decay factor with a value in the range of (0, 1); is the preset state transition matrix after introducing the transformation matrix T, and the transformed state transition matrix, and the transformed state transition matrix represents a diagonal matrix with the main diagonal elements being an arbitrary vector a;

[0083] Then, calculate the first innovation gain term K(k):

[0084]

[0085]

[0086]

[0087] where sat(.) represents the saturation function; and are respectively the preset smoothing layer parameter vectors; H1 = I 3×3 is the observation matrix of the full-rank block submatrix, and the full-rank block submatrix H1 = I 3×3 is linear time-invariant.

[0088] S1024: According to the target state prior estimate, the prior state estimate covariance matrix, the prior position measurement error, and the first innovation gain term, calculate the target state posterior estimate and the posterior state estimate covariance matrix of the current frame, and use the target state posterior estimate and the posterior state estimate covariance matrix of the current frame as the first-level target state estimate.

[0089] The posterior estimate of the target state at the current frame k and the state covariance matrix P p (k|k) are determined according to the following formula:

[0090]

[0091]

[0092] where, is the target state prior estimate, P p (k|k - 1) is the prior state estimate covariance matrix, e p (k|k - 1) is the prior position measurement error, and K(k) is the first innovation gain term; is the identity matrix, R p (k) is the first covariance, is the observation matrix.

[0093] The calculated posterior estimate and the state covariance matrix P of the posterior state p (k|k) are used as the results of the first-level target state estimation.

[0094] S103: The second measurement conversion module constructs a target range-radial velocity product pseudo-measurement using the target radial velocity measurement of the current frame Doppler radar.

[0095] Specifically, the second measurement conversion module constructs the range-radial velocity product pseudo-measurement using the target range measurement and the target radial velocity measurement of the Doppler radar, and calculates the second conversion bias μ η of the range-radial velocity product pseudo-measurement, the second conversion variance R η and the second conversion covariance R pη .

[0096] The range-radial velocity product pseudo-measurement is defined according to the following formula:

[0097]

[0098] where r m (k) is the range, is the target radial velocity measurement.

[0099] Next, calculate the second conversion bias μ η , the second conversion variance R η and the second conversion covariance R pη of the range-radial velocity product pseudo-measurement respectively:

[0100]

[0101]

[0102]

[0103] where ρ represents the correlation coefficient between the range observation error and the radial velocity observation error, σ represents the standard deviation of the observation error in each observation dimension, σ r represents the standard deviation of the range observation error, represents the standard deviation of the radial velocity observation error, represents the standard deviation of the pitch angle observation error, σ θ represents the standard deviation of the azimuth angle observation error.

[0104] S104: The second state estimator updates the first-stage target state estimate based on the target range-radial velocity product pseudo-measurement to obtain the final posterior state estimate of the target in the current frame.

[0105] This step specifically includes the following sub-steps:

[0106] S1041: Perform pre-whitening processing on the range-radial velocity product pseudo-measurement to obtain the pre-whitened pseudo-measurement, the bias of the pre-whitened pseudo-measurement, and the variance of the pre-whitened pseudo-measurement;

[0107] Specifically, the process of performing pre-whitening processing on the range-radial velocity product pseudo-measurement includes:

[0108] First, calculate the pre-whitening coefficient matrix L(k):

[0109] L(k) = -R ηp (k)(R P (k)) -1 = [L1(k) L2(k) L3(k)]

[0110] In the formula, R ηp (k) is the transpose of the second transformation covariance R pη (k), and R p (k) is the first covariance.

[0111] Next, according to the pre-whitening coefficient matrix, calculate the pre-whitened pseudo-measurement ε c (k), the transformation bias μ ε (k) of the pre-whitened pseudo-measurement, and the variance R ε (k) of the pre-whitened pseudo-measurement:

[0112]

[0113] Among them, the vector is the linear position measurement, and the scalar η c (k) is the target range-radial velocity product pseudo-measurement;

[0114] Find the transformation bias μ ε (k) of the pre-whitened pseudo-measurement:

[0115] μ ε (k) = L(k)μ p (k) + μ η (k)

[0116] Among them, the vector μ p (k) is the first transformation bias, and the scalar μ η (k) is the second transformation bias.

[0117] Find the variance R of the pre-whitened pseudo-measurement ε (k):

[0118] R ε (k)=R η (k)-(R pη (k)) T (R P (k)) -1 R pη (k)

[0119] where R p (k) is the first covariance, R η (k) is the second conversion variance, R pη (k) is the second conversion covariance.

[0120] S1042: Use the first-level target state estimate, the pre-whitened pseudo-measurement, the deviation of the pre-whitened pseudo-measurement, and the variance of the pre-whitened pseudo-measurement as inputs to construct a locally approximated linear minimum mean square error estimator to update the first-level target state estimate and obtain the final posterior state estimate of the current frame target.

[0121] This step specifically includes:

[0122] K ε (k)=P p (k|k)H ε (k) T [H ε (k)P p (k|k)H ε (k) T +R ε (k)] -1

[0123]

[0124]

[0125] P(k|k)=[I-K ε (k)H ε (k)]P p (k|k)[I-K ε (k)H ε (k)] T +K ε (k)R ε (k)K ε (k) T

[0126] where the non-linear function is defined as follows:

[0127]

[0128] Matrix H ε is a non - linear function of the Jacobian matrix, defined as:

[0129]

[0130] Take the corrected posterior state estimate and the corrected state covariance matrix P(k|k) as the final posterior state estimate of the target for the current frame k.

[0131] The present invention first converts the non - linear target position measurement of the Doppler radar into a linear position measurement in the Cartesian coordinate system, and then a generalized smooth variable - structure filter in linear form can be constructed to achieve a robust estimation of the target state. And by further updating and correcting the first - stage target state estimate of the output, the robust state - estimation performance under the condition of model uncertainty is ensured.

[0132] Secondly, this method uses the state - space sequential estimation method to divide the observation vector of the Doppler radar into two parts, including the target position measurement and the target radial velocity measurement, and sequentially updates the state estimate with these two parts of measurements respectively, solving the under - determined problems of the non - linear model of the Doppler radar and the non - full - rank observation matrix faced by the existing smooth variable - structure filtering methods. Furthermore, the radial velocity measurement information of the Doppler radar can be fully utilized to effectively improve the target state - estimation accuracy of the smooth variable - structure filter. Therefore, the method of the present invention makes more effective use of the Doppler velocity information than the existing smooth variable - structure filtering methods, thus significantly improving the accuracy of target state tracking.

[0133] This embodiment also provides a simulation scenario to conduct a comparative simulation experiment on the sequential smooth variable - structure filtering method (SSVSF), the extended Kalman filtering method (EKF), the smooth variable - structure filtering method using only position measurements (PO - SVSF), and the smooth variable - structure filtering method with local linearization of the Jacobian matrix (SVSF) proposed by the present invention.

[0134] Maneuvering target tracking is a typical scenario of the model - uncertainty problem. Therefore, a simulation scenario for the tracking and filtering of a single maneuvering target by a 2 - D Doppler millimeter - wave radar in an autonomous driving scenario is made, and the radar is established at the origin of coordinates.

[0135] Refer to Figure 2 , Figure 2 is the target maneuvering trajectory diagram in the simulation environment, Figure 2 in which the abscissa x and the ordinate y form a two - dimensional plane. The solid dots in the figure are the Doppler radars in the simulation environment, and the dashed lines in the figure are the maneuvering trajectories of the target objects.

[0136] Generate simulated radar target simulation data using the simulation parameters in Table 1 below. The target state simulation scalar is defined as the position and velocity within the X-Y coordinate axes, i.e., x = [x y v x v y T ; The radar measurement variable is defined Use the constant velocity motion model (CV) to estimate the target state, i.e.:

[0137]

[0138] The measurement matrix of the filter in the first-stage processing of the sequential smoothing variable structure filtering method is:

[0139]

[0140] Table 1 Radar target tracking scenario parameters

[0141]

[0142]

[0143] Conduct simulation experiments according to the sequential smoothing variable structure filtering method (SSVSF), extended Kalman filtering method (EKF), smoothing variable structure filtering method using only position measurements (PO-SVSF), and smoothing variable structure filtering method with local linearization of the Jacobian matrix (SVSF) provided in this embodiment, and use the average results of 2000 Monte Carlo experiments as comparison data; Table 2 below gives the root mean square error (RMSE) of 2000 Monte Carlo experiments.

[0144] Table 2 Root mean square error of state estimation of targets with uncertain motion models

[0145]

[0146] Obviously, it can be seen from Table 2 that the sequential smoothing variable structure filtering method (SSVSF) proposed in this embodiment has the smallest root mean square error of target state estimation under the condition of uncertain motion models.

[0147] Figure 3 Shows the variation of the coordinate error of target state estimation for each method with the tracking time, Figure 4 Shows the variation of the velocity error of target state estimation for each method with the tracking time.

[0148] From Figure 3 and Figure 4 ​It can be seen that the proposed sequential smooth variable structure filtering (SSVSF) method of the present invention achieves the best tracking accuracy. Compared with the extended Kalman filter (EKF), the SSVSF method has more robust tracking performance during the maneuvering motion of the target. For example, at 10 - 15 s and 23 - 28 s, the peak coordinate error is reduced by more than half; compared with the PO-SVSF that only uses position measurements, the proposed SSVSF can effectively utilize Doppler velocity information and greatly improve the target state estimation accuracy; while the local linearized SVSF faces the underdetermined problem caused by the non-full-rank observation matrix, and the target state estimation performance, especially the velocity estimation performance, is the worst.

[0149] Therefore, through simulation experiments, the method of the present invention solves the problem of the non-linear underdetermined observation model of Doppler radar. At the same time, under the condition of uncertain target motion models, it can make full use of Doppler velocity measurement information to achieve more robust and accurate radar target state estimation.

[0150] An embodiment of this application provides a Doppler radar sequential smooth variable structure filtering device. The filtering device includes a first measurement conversion module 100, a first state estimator 200, a second measurement conversion module 300, and a second state estimator 400, where:

[0151] The first measurement conversion module 100 is used to convert the non-linear target position measurement of the current frame Doppler radar into a linear position measurement in the Cartesian coordinate system;

[0152] The first state estimator 200 is used to obtain a first-level target state estimation by using a generalized smooth variable structure filter and the linear position measurement in the Cartesian coordinate system;

[0153] The second measurement conversion module 300 is used to construct a target range-radial velocity product pseudo-measurement by using the target radial velocity measurement of the current frame Doppler radar;

[0154] The second state estimator 400 is used to update the first-level target state estimation based on the target range-radial velocity product pseudo-measurement to obtain the final posterior state estimation of the target for the current frame.

[0155] The first measurement conversion module is used to convert the non-linear target position measurement vector in the three-dimensional spherical coordinates or two-dimensional polar coordinates of the Doppler radar into the linear position measurement vector in the Cartesian coordinate system by using an unbiased measurement conversion method, and calculate the first conversion deviation and the first conversion covariance.

[0156] The first state estimator includes:

[0157] A first calculation unit for calculating a prior estimate of the target state, a prior estimate of radar measurements, and a prior state estimate covariance matrix for the current frame;

[0158] A second calculation unit for calculating a prior position measurement error based on the linear position measurement of the current frame, the first conversion deviation, the first conversion covariance, and the prior estimate of radar measurements;

[0159] A third calculation unit for calculating a first innovation gain term based on the prior position measurement error;

[0160] A fourth calculation unit for calculating a posterior estimate of the target state and a posterior state estimate covariance matrix for the current frame based on the prior estimate of the target state, the prior state estimate covariance matrix, the prior position measurement error, and the first innovation gain term, and taking the posterior estimate of the target state and the posterior state estimate covariance matrix for the current frame as the first-level target state estimate.

[0161] The second measurement conversion module is used to construct the range-radial velocity product pseudo-measurement by using the target range measurement and the target radial velocity measurement of the Doppler radar, and calculate the second conversion deviation, the second conversion variance, and the second conversion covariance of the range-radial velocity product pseudo-measurement by using an unbiased conversion method.

[0162] The second state estimator includes:

[0163] A pre-whitening processing unit for performing pre-whitening processing on the range-radial velocity product pseudo-measurement to obtain a pre-whitened pseudo-measurement, the deviation of the pre-whitened pseudo-measurement, and the variance of the pre-whitened pseudo-measurement;

[0164] A posterior state estimation unit for using the first-level target state estimate, the pre-whitened pseudo-measurement, the deviation of the pre-whitened pseudo-measurement, and the variance of the pre-whitened pseudo-measurement as inputs to construct a locally approximated linear minimum mean square error estimator to update the first-level target state estimate and obtain the final posterior state estimate of the current frame target.

[0165] Each embodiment in this specification is described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other.

[0166] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, devices, or computer program products. Therefore, the embodiments of the present application can take the form of completely hardware embodiments, completely software embodiments, or embodiments combining software and hardware aspects. Moreover, the embodiments of the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0167] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of methods, terminal devices (devices), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing terminal devices to generate a machine, such that the instructions executed by the processors of the computer or other programmable data processing terminal devices generate a device for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0168] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device that implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0169] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, such that a series of operation steps are executed on the computer or other programmable terminal device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable terminal device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0170] Although the preferred embodiments of the embodiments of the present application have been described, those skilled in the art can make additional changes and modifications to these embodiments once they know the basic creative concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present application.

[0171] Finally, it should also be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or terminal device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or terminal device comprising the said element.

[0172] In this text, specific examples are used to elaborate on the principles and implementation manners of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A sequential smoothing variable structure filtering method for Doppler radar, characterized in that, The method includes the following steps: Using a first measurement conversion module, convert the non-linear target position measurement vector of the current frame Doppler radar into a linear position measurement vector in the Cartesian coordinate system; Using a first state estimator, obtain a first-level target state estimate by using a generalized smooth variable structure filter and the linear position measurement vector in the Cartesian coordinate system; Using a second measurement conversion module, construct a target range-radial velocity product pseudo-measurement by using the target radial velocity measurement of the current frame Doppler radar; Using a second state estimator, update the first-level target state estimate based on the target range-radial velocity product pseudo-measurement to obtain the final posterior state estimate of the current frame target.

2. The filtering method according to claim 1, wherein The first measurement conversion module uses an unbiased measurement conversion method to convert the non-linear target position measurement vector in the three-dimensional spherical coordinates or two-dimensional polar coordinates of the Doppler radar into the linear position measurement vector in the Cartesian coordinate system, and calculates a first conversion deviation and a first conversion covariance.

3. The filtering method according to claim 2, wherein The first state estimator, using a generalized smooth variable structure filter and the linear position measurement vector in the Cartesian coordinate system, obtains a first-level target state estimate, including the following steps: Calculate the prior estimate of the target state, the prior estimate of the radar measurement, and the prior state estimate covariance matrix of the current frame; Calculate the prior position measurement error according to the linear position measurement vector of the current frame, the first conversion deviation, the first conversion covariance, and the prior estimate of the radar measurement; Calculate a first innovation gain term according to the prior position measurement error; Calculate the posterior estimate of the target state and the posterior state estimate covariance matrix of the current frame according to the prior estimate of the target state, the prior state estimate covariance matrix, the prior position measurement error, and the first innovation gain term, and use the posterior estimate of the target state and the posterior state estimate covariance matrix of the current frame as the first-level target state estimate.

4. The filtering method according to any one of claims 1-3, characterized in that, The second measurement conversion module uses the target range measurement and the target radial velocity measurement of the Doppler radar to construct the range-radial velocity product pseudo-measurement, and uses an unbiased conversion method to calculate a second conversion deviation, a second conversion variance, and a second conversion covariance of the range-radial velocity product pseudo-measurement.

5. The filtering method according to claim 4, characterized in that The second state estimator, based on the target range-radial velocity product pseudo-measurement, updates the first-level target state estimate to obtain the final posterior state estimate of the current frame target, including the following steps: Perform pre-whitening processing on the range-radial velocity product pseudo-measurement to obtain a pre-whitened pseudo-measurement, the deviation of the pre-whitened pseudo-measurement, and the variance of the pre-whitened pseudo-measurement; Use the first-level target state estimate, the pre-whitened pseudo-measurement, the deviation of the pre-whitened pseudo-measurement, and the variance of the pre-whitened pseudo-measurement as inputs to construct a locally approximated linear minimum mean square error estimator to update the first-level target state estimate to obtain the final posterior state estimate of the current frame target.

6. A sequential smoothing variable structure filtering device for a Doppler radar, characterized in that, The filtering device includes a first measurement conversion module, a first state estimator, a second measurement conversion module, and a second state estimator, where: The first measurement conversion module is used to convert the non-linear target position measurement vector of the current frame Doppler radar into a linear position measurement vector in the Cartesian coordinate system; The first state estimator is used to obtain a first-level target state estimate by using a generalized smooth variable structure filter and the linear position measurement vector in the Cartesian coordinate system; The second measurement conversion module is used to construct a target range-radial velocity product pseudo-measurement by using the target radial velocity measurement of the current frame Doppler radar; The second state estimator is used to update the first-level target state estimate based on the target range-radial velocity product pseudo-measurement to obtain the final posterior state estimate of the current frame target.

7. The filtering device according to claim 6, wherein The first measurement conversion module uses an unbiased measurement conversion method to convert the non-linear target position measurement vector in the three-dimensional spherical coordinates or two-dimensional polar coordinates of the Doppler radar into the linear position measurement vector in the Cartesian coordinate system, and calculates the first conversion deviation and the first conversion covariance.

8. The filtering device according to claim 7, characterized in that, The first state estimator includes: A first calculation unit for calculating the prior estimate of the target state of the current frame, the prior estimate of the radar measurement, and the prior state estimate covariance matrix; A second calculation unit for calculating the prior position measurement error according to the linear position measurement vector of the current frame, the first conversion deviation, the first conversion covariance, and the prior estimate of the radar measurement; A third calculation unit for calculating a first innovation gain term according to the prior position measurement error; A fourth calculation unit for calculating the posterior estimate of the target state and the posterior state estimate covariance matrix of the current frame according to the prior estimate of the target state, the prior state estimate covariance matrix, the prior position measurement error, and the first innovation gain term, and taking the posterior estimate of the target state and the posterior state estimate covariance matrix of the current frame as the first-level target state estimate.

9. The filtering device according to any one of claims 6-8, characterized in that, The second measurement conversion module is used to construct the range-radial velocity product pseudo-measurement by using the target range measurement and the target radial velocity measurement of the Doppler radar, and calculates the second conversion deviation, the second conversion variance, and the second conversion covariance of the range-radial velocity product pseudo-measurement by using an unbiased conversion method.

10. The filtering device according to claim 9, characterized in that, The second state estimator includes: A pre-whitening processing unit for performing pre-whitening processing on the range-radial velocity product pseudo-measurement to obtain a pre-whitened pseudo-measurement, the deviation of the pre-whitened pseudo-measurement, and the variance of the pre-whitened pseudo-measurement; A posterior state estimation unit for using the first-level target state estimate, the pre-whitened pseudo-measurement, the deviation of the pre-whitened pseudo-measurement, and the variance of the pre-whitened pseudo-measurement as inputs to construct a locally approximated linear minimum mean square error estimator to update the first-level target state estimate to obtain the final posterior state estimate of the current frame target.

Images