Robust Tracking Method for Small and Weak Multiple Targets Under Dynamic Interference Based on Q-IMM-MHT.

Area of Science:

• Computer Vision
• Robotics
• Signal Processing

Background:

• Traditional multi-target tracking struggles with clutter and trajectory interruptions, leading to poor accuracy and robustness.
• Existing methods like UKF, IMM, and CIMM have limitations in complex, dynamic environments.

Purpose of the Study:

• To develop an advanced multi-target tracking algorithm, Q-IMM-MHT, for improved performance in challenging conditions.
• To enhance data association accuracy and system robustness through adaptive model switching and anomaly detection.

Main Methods:

• Integration of Multiple Hypothesis Tracking (MHT) with Interactive Multiple Model (IMM).
• Implementation of a Q-learning-based adaptive model switching strategy for dynamic motion pattern adjustments.
• Utilization of Support Vector Machines (SVMs) for anomaly detection and trajectory recovery.

Main Results:

• Reduced Root Mean Square Error (RMSE) to 0.74 pixels (position) and 0.04 pixels/frame (velocity) under high noise.
• Achieved at least 10.84% and 42.86% RMSE reduction compared to UKF, IMM, and CIMM.
• Maintained >46.3% association accuracy after 30 frames of interruption in cluttered environments.

Conclusions:

• Q-IMM-MHT demonstrates significant performance improvements in multi-target tracking within complex environments.
• The algorithm enhances tracking accuracy, stability, and robustness, showing considerable application potential.