Radar track Kalman filtering method and system for optimizing gating circulation unit parameters

Abstract

The invention provides a radar track Kalman filtering method and system for gating cycle unit parameter optimization. The method comprises the following steps: step 1, constructing a Kalman-gating cycle unit track filtering calculation framework; step 2, collecting radar target track measurement data and corresponding true value sensor data to construct a training data set, training a Kalman-gated cycle unit in a framework, optimizing Kalman filtering process noise covariance estimation, observing noise covariance estimation, and learning gated cycle unit network parameters of real-time optimization gain; and step 3, based on an optimization learning result in the step 2, carrying out track filtering on radar real-time measurement information through a Kalman-gating cycle unit track filtering calculation framework. According to the invention, the accuracy and robustness of radar track state estimation are improved.

Description

Technical Field

[0001] This invention belongs to the field of radar data processing, specifically relating to a radar track Kalman filtering method and system for optimizing gated cyclic unit parameters. Background Technology

[0002] Radar target tracking is a crucial component of radar detection systems. Kalman filtering is a commonly used and effective method for radar target tracking, representing the optimal estimation model for linear systems under the assumption of independent and identically distributed Gaussian noise. However, in radar target tracking systems, factors such as noise, interference, coordinate transformation, and target maneuvering often violate the assumptions of traditional Kalman filtering, rendering it a suboptimal filter and leading to non-convergence or even divergence. Therefore, a nonlinear filtering model parameter optimization method is urgently needed to improve the accuracy and robustness of radar track state estimation.

[0003] Neural networks are an effective method for optimizing nonlinear model parameters. Recurrent Neural Networks (RNNs) are highly effective for predicting time-series data, but they are prone to gradient vanishing or exploding, making them unsuitable for processing long sequences. Long Short-Term Memory (LSTM) networks effectively solve the gradient vanishing or exploding problems, but their complex structure and numerous parameters make them suitable for handling complex time-series data. Gated Recurrent Units (GRUs) are a simplification of LSTM networks, offering a simpler structure and fewer parameters while maintaining performance. Radar target tracking systems have simple data structures and high real-time requirements; therefore, GRUs are more suitable for optimizing the parameters of nonlinear filtering models in radar target tracking systems.

[0004] Guilin University of Electronic Technology disclosed a trajectory prediction method based on radar multi-target tracking in its invention patent application "A Trajectory Prediction Method Based on Radar Multi-Target Tracking" (Publication No.: CN118915071A). This method only predicts the vehicle's intention and future trajectory in the current step through a BiLSTM network, without optimizing the parameters of the nonlinear filtering model.

[0005] Guangzhou Vision Intelligent Technology Co., Ltd. and South China University of Technology disclosed a three-dimensional target tracking method based on Kalman filtering and LSTM in their invention patent application "A Three-Dimensional Target Tracking Method Based on Kalman Filtering and LSTM" (Publication No.: CN111161325B). The method only uses the LSTM network to predict the target position and does not optimize the parameters of the nonlinear filtering model.

[0006] Nanjing Thunder Information Technology Co., Ltd. disclosed a method for overcoming the divergence of radar extended Kalman track filtering in its invention patent application "A method for overcoming the divergence of radar extended Kalman track filtering" (publication number: CN112986977B). This method requires manually setting the modified prediction covariance matrix and does not have adaptive characteristics.

[0007] Sichuan Jiuzhou Air Traffic Control Technology Co., Ltd. disclosed a secondary radar adaptive trajectory filtering method in its invention patent application "A Secondary Radar Adaptive Track Filtering Method" (Publication No.: CN112731374A). This method judges non-maneuvering or maneuvering targets by manually setting preset conditions, which is not intelligent enough. Summary of the Invention

[0008] This invention provides a radar track Kalman filtering method and system with optimized gated cyclic unit parameters to solve the problems in the background technology and improve the accuracy and robustness of radar track state estimation.

[0009] The technical solution to achieve the purpose of this invention is as follows:

[0010] A Kalman filtering method for radar tracks with optimized gated cyclic unit parameters includes:

[0011] Step 1: Construct a Kalman-gated cyclic unit trajectory filtering calculation framework;

[0012] Step 2: Collect radar target trajectory measurement data set and corresponding true sensor data Building a training dataset Train a Kalman-gated recurrent unit to optimize noise covariance estimation in the Kalman filtering process. Observation noise covariance estimation Learn to optimize gain in real time Gated cyclic unit network parameters;

[0013] Step 3: Utilize optimized process noise covariance estimation Observation noise covariance estimation Real-time optimization of Kalman gain with trained gated recurrent units The radar real-time measurement information is filtered based on the Kalman-gated cyclic unit trajectory filtering calculation framework constructed in step 1.

[0014] Preferably, step 1 specifically includes:

[0015] Step 1-1: Set up and initialize the Kalman filter target motion state model;

[0016] Step 1-2: Perform Kalman prediction based on the state estimate X(t|t) and covariance matrix P(t|t) at time t to obtain the state prediction X(t+1|t) and covariance prediction P(t+1|t), and calculate the gain K(t+1) at time t+1.

[0017] Steps 1-3: Calculate the innovation v(t+1) based on the radar measurement data z(t+1) at time t+1;

[0018] Steps 1-4: Use the gain K(t+1) as the hidden state input of the gated loop unit, and the innovation v(t+1) as the current time input of the gated loop unit. Use the hidden layer output after processing by the gated loop unit as the gain for Kalman state update. The state estimate X(t+1|t+1) and covariance P(t+1|t+1) at time t+1 are calculated accordingly.

[0019] Steps 1-5: Repeat steps 1-2 to 1-4 in a time-series loop to perform radar track filtering at each time point.

[0020] Preferably, step 1-2 calculates the gain K(t+1) at time t+1 as follows:

[0021] X(t+1|t)=F(t+1)X(t|t)

[0022] P(t+1|t)=F(t+1)P(t|t)F(t+1) T +Q

[0023] K(t+1)=P(t+1|t)H(t+1) T (H(t+1)P(t+1|t)H(t+1) T +R) -1

[0024] Where Q is the process noise covariance, R is the observation noise covariance, and F(t+1) is the value of F. T H(t+1) is the transpose of the state transition matrix F(t+1) at time t+1. T It is the transpose of the observation matrix H(t+1) at time t+1.

[0025] Preferably, steps 1-3 calculate the innovation v(t+1) as follows:

[0026] v(t+1)=z(t+1)-H(t+1)X(t+1|t).

[0027] Preferably, steps 1-4 calculate the state estimate X(t+1|t+1) and covariance P(t+1|t+1) at time t+1 as follows:

[0028]

[0029] Preferably, a training dataset is constructed. for:

[0030]

[0031] Where t represents the measurement time and i represents the target track number. For radar target trajectory measurement dataset, For corresponding true value sensor data;

[0032] Preferably, the Kalman-gated recurrent unit is trained to optimize the noise covariance estimation of the Kalman filtering process. Observation noise covariance estimation Learn to optimize gain in real time The parameters of the gated recurrent unit network specifically include:

[0033] Step 2-1: Based on the training dataset Initialize the Kalman filter initial state for each target's radar measurement data. R, Q,

[0034] Step 2-2: Estimation of radar track state at time t Covariance Perform a prediction step and calculate the gain at time t+1. As the hidden state input of the gated loop unit, combined with the radar measurement data at time t+1. Calculate new information As the current input of the gated loop unit;

[0035] Steps 2-3: The hidden state output calculated by the gated loop unit is used as the Kalman filter gain. Perform a state update to obtain the state estimate at time t+1. Covariance

[0036] Steps 2-4: Set the network parameters of Q, R, and the gated recurrent unit to trainable parameters, based on the training samples. and Design the loss function, and set the error function as follows:

[0037]

[0038] Where δ is a hyperparameter, set Conduct training;

[0039] Step 2-5: Repeat steps 2-2 to 2-4 until all time points T and target k have been trained, and optimize the noise covariance estimation of the Kalman filter process. Observation noise covariance estimation And to obtain optimized gain in real time The parameters of the gated cyclic unit network.

[0040] Preferably, the loss function is designed as follows:

[0041]

[0042] Where δ is a hyperparameter, set Conduct training.

[0043] A Kalman filter system for radar tracks with gated cyclic unit parameter optimization includes:

[0044] The building block is used to construct the Kalman-gated cyclic unit trajectory filtering calculation framework;

[0045] The training unit collects radar target trajectory measurement data and corresponding ground truth sensor data to construct a training dataset. In the training framework, the Kalman-gated recurrent unit optimizes the noise covariance estimation of the Kalman filtering process and the observation noise covariance estimation, and learns the network parameters of the gated recurrent unit to optimize the gain in real time.

[0046] The filtering unit, based on the optimized learning results of the training unit, performs track filtering on the real-time radar measurement information through the Kalman-gated cyclic unit track filtering calculation framework.

[0047] A computer storage medium storing an executable program, the executable program being executed by a processor to perform the steps of the radar track Kalman filtering method described above.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows: by using radar measurement data and true sensor data to construct training data to train the Kalman-gated recurrent unit, the noise estimation of the Kalman filtering process can be effectively optimized. Observation noise estimation and gain The parameters are designed to be suitable for nonlinear tracking systems, significantly improving the accuracy and robustness of radar track state estimation; and a loss function is designed to improve the accuracy of Kalman-gated recurrent unit training. Attached Figure Description

[0049] Figure 1 This is a flowchart of the radar track Kalman filtering method of the present invention. Detailed Implementation

[0050] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0051] This invention proposes a Kalman filtering method for radar tracks with optimized gated cyclic unit parameters. It constructs a Kalman-gated cyclic unit track filtering calculation framework and collects radar target track measurement datasets. and corresponding true sensor data Building a training dataset Training a Kalman-gated recurrent unit can effectively optimize noise estimation in the Kalman filtering process. Observation noise estimation and gain These parameters are suitable for nonlinear tracking systems, significantly improving the accuracy and robustness of radar track state estimation. The preferred implementation process is shown in the appendix. Figure 1 The preferred embodiment includes the following steps:

[0052] Step 1: Construct a Kalman-gated cyclic unit trajectory filtering calculation framework, including:

[0053] Step 1-1: Establish and initialize the Kalman filter constant acceleration target motion state model X(t|t) and state transition matrix F(t);

[0054]

[0055]

[0056] Steps 1-2: Perform Kalman prediction based on the state estimate X(t|t) and covariance matrix P(t|t) at time t to obtain the state prediction X(t+1|t) and covariance prediction P(t+1|t), and calculate the gain K(t+1) at time t+1.

[0057] X(t+1|t)=F(t+1)X(t|t)

[0058] P(t+1|t)=F(t+1)P(t|t)F(t+1) T +Q

[0059] K(t+1)=P(t+1|t)H(t+1) T (H(t+1)P(t+1|t)H(t+1) T +R) -1

[0060] Where Q is the process noise covariance, R is the observation noise covariance, and F(t+1) is the value of F. T H(t+1) is the transpose of the state transition matrix F(t+1) at time t+1. T It is the transpose of the observation matrix H(t+1) at time t+1;

[0061] Steps 1-3: Calculate the innovation v(t+1) based on the radar measurement data z(t+1) at time t+1:

[0062] v(t+1)=z(t+1)-H(t+1)X(t+1|t)

[0063] Steps 1-4: Use the gain K(t+1) as the hidden state input of the gated loop unit, and the innovation v(t+1) as the current time input of the gated loop unit. Use the hidden layer output after processing by the gated loop unit as the gain for Kalman state update. Calculate the state estimate X(t+1|t+1) and covariance P(t+1|t+1) at time t+1:

[0064]

[0065] Steps 1-5: Repeat steps 1-2 to 1-4 in a time-series loop to perform radar track filtering at each time point.

[0066] Step 2: Collect radar target trajectory measurement data set and corresponding GPS data Building a training dataset

[0067]

[0068] Where t represents the measurement time and i represents the target track number;

[0069] Train a Kalman-gated recurrent unit to optimize noise covariance estimation in the Kalman filtering process. Observation noise covariance estimation Learn to optimize gain in real time The parameters of the gated recurrent unit network include:

[0070] Step 2-1: Based on the training dataset Initialize the Kalman filter initial state for each target's radar measurement data. R, Q,

[0071] Step 2-2: Estimation of radar track state at time t Covariance Perform a prediction step and calculate the gain at time t+1. As the hidden state input of the gated loop unit, combined with the radar measurement data at time t+1. Calculate new information As the current input of the gated loop unit;

[0072] Steps 2-3: The hidden state output calculated by the gated loop unit is used as the Kalman filter gain. Perform a state update to obtain the state estimate at time t+1. Covariance

[0073] Steps 2-4: Set the network parameters of Q, R, and the gated recurrent unit to trainable parameters, based on the training samples. and Calculate the error function:

[0074]

[0075] Where δ is a hyperparameter, set Set hyperparameters such as learning rate, batch size, number of training iterations, and Dropout regularization, and select the Adam optimization method for parameter training;

[0076] Step 2-5: Repeat steps 2-2 to 2-4 until all time points T and target k have been trained, and optimize the noise covariance estimation of the Kalman filter process. Observation noise covariance estimation And to obtain optimized gain in real time Gated cyclic unit network parameters;

[0077] Step 3: Utilize optimized process noise covariance estimation Observation noise covariance estimation Real-time optimization of Kalman gain with trained gated recurrent units The radar real-time measurement information is filtered based on the Kalman-gated cyclic unit trajectory filtering calculation framework constructed in step 1.

[0078] This embodiment also provides a radar track Kalman filter system with gated cyclic unit parameter optimization, including:

[0079] The building block is used to construct the Kalman-gated cyclic unit trajectory filtering calculation framework;

[0080] The training unit collects radar target trajectory measurement data and corresponding ground truth sensor data to construct a training dataset. In the training framework, the Kalman-gated recurrent unit optimizes the noise covariance estimation of the Kalman filtering process and the observation noise covariance estimation, and learns the network parameters of the gated recurrent unit to optimize the gain in real time.

[0081] The filtering unit, based on the optimized learning results of the training unit, performs track filtering on the real-time radar measurement information through the Kalman-gated cyclic unit track filtering calculation framework.

[0082] This embodiment also provides a computer storage medium storing an executable program, which is executed by a processor to perform the steps of the radar track Kalman filtering method described herein.

[0083] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0084] Obviously, those skilled in the art can make various modifications and variations to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. Thus, if these modifications and variations to the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention also intends to include these modifications and variations.

Claims

1. A Kalman filtering method for radar tracks with optimized gated cyclic unit parameters, characterized in that, include: Step 1: Construct a Kalman-gated cyclic unit trajectory filtering calculation framework; Step 2: Collect radar target trajectory measurement data and corresponding ground truth sensor data to construct a training dataset. Train the Kalman-gated recurrent unit in the framework, optimize the noise covariance estimation of the Kalman filtering process and the observation noise covariance estimation, and learn the network parameters of the gated recurrent unit to optimize the gain in real time. Step 3: Based on the optimization learning results of Step 2, track filtering is performed on the real-time radar measurement information using the Kalman-gated cyclic unit track filtering calculation framework.

2. The radar track Kalman filtering method for gated cyclic unit parameter optimization according to claim 1, characterized in that, Step 1, which constructs the Kalman-gated cyclic unit trajectory filtering calculation framework, specifically includes: Step 1-1: Initialize the Kalman filter target motion state model; Step 1-2: Perform Kalman prediction based on the state estimate X(t|t) and covariance matrix P(t|t) at time t to obtain the state prediction X(t+1|t) and covariance prediction P(t+1|t), and calculate the gain K(t+1) at time t+1. Steps 1-3: Calculate the innovation v(t+1) based on the radar measurement data z(t+1) at time t+1; Steps 1-4: Use the gain K(t+1) as the hidden state input of the gated loop unit, and the innovation v(t+1) as the current time input of the gated loop unit. Use the hidden layer output after processing by the gated loop unit as the gain for Kalman state update. The state estimate X(t+1|t+1) and covariance P(t+1|t+1) at time t+1 are calculated accordingly. Steps 1-5: Repeat steps 1-2 to 1-4 in a time-sequence loop to perform radar track filtering at each time point.

3. The radar track Kalman filtering method for gated cyclic unit parameter optimization according to claim 2, characterized in that, In step 1-2, the gain K(t+1) at time t+1 is calculated as follows: X(t+1|t)=F(t+1)X(t|t) P(t+1|t)=F(t+1)P(t|t)F(t+1) T +Q K(t+1)=P(t+1|t)H(t+1) T (H(t+1)P(t+1|t)H(t+1) T +R) -1 Where Q is the process noise covariance, R is the observation noise covariance, and F(t+1) is the value of F. T H(t+1) is the transpose of the state transition matrix F(t+1) at time t+1. T It is the transpose of the observation matrix H(t+1) at time t+1.

4. The radar track Kalman filtering method for gated cyclic unit parameter optimization according to claim 2, characterized in that, Steps 1-3 calculate the new information v(t+1) as follows: v(t+1)=z(t+1)-H(t+1)X(t+1|t).

5. The radar track Kalman filtering method for gated cyclic unit parameter optimization according to claim 2, characterized in that, Steps 1-4 calculate the state estimate X(t+1|t+1) and covariance P(t+1|t+1) at time t+1 as follows:

6. The radar track Kalman filtering method for gated cyclic unit parameter optimization according to claim 1, characterized in that, Step 2: Construct the training dataset as follows: Where t represents the measurement time and i represents the target track number. For radar target trajectory measurement dataset, This corresponds to the true value sensor data.

7. The radar track Kalman filtering method for gated cyclic unit parameter optimization according to claim 1, characterized in that, Step 2 involves training the Kalman-gated recurrent units in the framework, optimizing the noise covariance estimation of the Kalman filtering process and the observation noise covariance estimation, and learning the gated recurrent unit network parameters for real-time gain optimization. Specifically, this includes: Step 2-1: Based on the training dataset Initialize the Kalman filter initial state with the radar measurement data of each target; Step 2-2: Estimation of radar track state at time t Covariance Perform a prediction step and calculate the gain at time t+1. As the hidden state input of the gated loop unit, combined with the radar measurement data at time t+1. Calculate new information As the current input of the gated loop unit; Steps 2-3: The hidden state output calculated by the gated loop unit is used as the Kalman filter gain. Perform a state update to obtain the state estimate at time t+1. Covariance Steps 2-4: Set the process noise covariance Q, observation noise covariance R, and network parameters of the gated recurrent unit as trainable parameters, and design a loss function for training; Step 2-5: Repeat steps 2-2 to 2-4 until training is complete, obtaining the optimized Kalman filter process noise covariance estimate. Observation noise covariance estimation And to obtain optimized gain in real time The parameters of the gated cyclic unit network.

8. The radar track Kalman filtering method for gated cyclic unit parameter optimization according to claim 7, characterized in that, The loss function is: Where δ is a hyperparameter.

9. A radar track Kalman filtering system implementing the radar track Kalman filtering method according to any one of claims 1-8, characterized in that, include: The building block is used to construct the Kalman-gated cyclic unit trajectory filtering calculation framework; The training unit collects radar target trajectory measurement data and corresponding ground truth sensor data to construct a training dataset. In the training framework, the Kalman-gated recurrent unit optimizes the noise covariance estimation of the Kalman filtering process and the observation noise covariance estimation, and learns the network parameters of the gated recurrent unit to optimize the gain in real time. The filtering unit, based on the optimized learning results of the training unit, performs track filtering on the real-time radar measurement information through the Kalman-gated cyclic unit track filtering calculation framework.

10. A computer storage medium, characterized in that, The computer storage medium stores an executable program, which is executed by a processor to implement the steps of the radar track Kalman filtering method according to any one of claims 1-8.

Images