Multi-Maneuvering Target Tracking Based on a Gaussian Process.

Area of Science:

• Signal Processing
• Machine Learning
• Control Systems

Background:

• Traditional multi-model (MM) methods for multi-maneuvering target tracking (MMTT) struggle with model uncertainty and predefined limitations.
• Existing methods often assume simplified linear or nonlinear observation models, leading to performance degradation in real-world applications.

Purpose of the Study:

• To address the limitations of traditional MMTT methods by developing an innovative data-driven approach using Gaussian processes (GP).
• To enhance adaptability and flexibility in tracking complex and dynamic multi-target scenarios with unknown motion and observation models.

Main Methods:

• A non-parametric Gaussian Process (GP) approach is employed to learn an unlimited range of target motion and observation models.
• Learned models are integrated into the cubature Kalman probability hypothesis density (PHD) filter for accurate MMTT estimation.

Main Results:

• The proposed GP-based method effectively mitigates issues of model overload and mismatch inherent in traditional MMTT techniques.
• Simulation results demonstrate high-precision tracking performance in complex multi-maneuvering target scenarios, confirming the approach's effectiveness.

Conclusions:

• The Gaussian Process-based method offers a robust solution for MMTT under model uncertainty, outperforming traditional approaches.
• This learning-based algorithm significantly improves tracking accuracy and adaptability in challenging dynamic environments.