Reinforcement Learning-Based Data Association for Multiple Target Tracking in Clutter.

Area of Science:

• * Signal Processing
• * Artificial Intelligence
• * Control Systems

Background:

• * Data association is critical for multiple target tracking, determining sensor measurement-to-target correspondence.
• * Existing methods often struggle with accuracy and computational efficiency in dense clutter scenarios.

Purpose of the Study:

• * To introduce a novel data association method, RL-JPDA, leveraging reinforcement learning.
• * To enhance accuracy and reduce computational complexity in multi-target tracking amidst clutter.

Main Methods:

• * Employing reinforcement learning (RL) to effectively utilize measurement information.
• * Integrating target motion characteristics to improve association accuracy.
• * Comparative analysis against Global Nearest Neighbor (GNN), JPDA, Fuzzy Optimal, and Intuitionistic Fuzzy JPDA methods.

Main Results:

• * The proposed RL-JPDA method demonstrates significantly shorter execution times compared to benchmark methods.
• * Achieves effective and feasible target state estimation in dense clutter environments.
• * Maintains high association accuracy, outperforming traditional approaches.

Conclusions:

• * Reinforcement learning provides a powerful framework for advancing data association techniques.
• * The RL-JPDA method offers a computationally efficient and accurate solution for challenging multi-target tracking problems.