Noise-Adaption Extended Kalman Filter Based on Deep Deterministic Policy Gradient for Maneuvering Targets

Area of Science:

• Signal Processing
• Control Systems
• Robotics

Background:

• Maneuvering target tracking is crucial in various applications.
• Existing methods struggle to differentiate unknown maneuvers from measurement inaccuracies, impacting performance.
• Lack of robust distinction leads to low accuracy, poor robustness, and filter divergence.

Purpose of the Study:

• To propose a noise-adaptive extended Kalman filter for robust maneuvering target tracking with multiple sensors.
• To reliably distinguish between unknown maneuvers and inaccurate measurements.
• To improve the accuracy and robustness of target tracking algorithms.

Main Methods:

• Developed a noise-adaptive extended Kalman filter integrating Dempster-Shafer evidence theory for maneuver detection.
• Employed a Markovian decision process and deep deterministic policy gradient for adaptive process noise covariance estimation.
• Utilized recursive estimation for measurement noise covariance and a fusion algorithm for global estimation.

Main Results:

• The proposed filter effectively distinguishes unknown maneuvers from inaccurate measurements by fusing multi-sensor information.
• Adaptive estimation of process noise covariance and recursive measurement noise covariance improved filter performance.
• Simulation results demonstrated the feasibility and superiority of the proposed algorithm in two scenarios.

Conclusions:

• The novel noise-adaptive filter significantly enhances the robustness and accuracy of maneuvering target tracking.
• The Dempster-Shafer evidence theory based approach provides reliable distinction between maneuvers and measurement noise.
• The adaptive estimation strategies contribute to superior performance compared to existing methods.