Enhancing Target Tracking: A Novel Grid-Based Beetle Antennae Search Algorithm and Confusion-Aware Detection.

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Unmanned aerial vehicle (UAV) target tracking faces challenges like limited computation, real-time needs, and target confusion.
• Existing methods struggle with efficiency and robustness in complex tracking scenarios.

Purpose of the Study:

• To develop an efficient and robust multi-target detection and positioning system for UAVs.
• To enhance UAV tracking performance under resource constraints and target ambiguity.

Main Methods:

• Proposed a grid-based beetle antennae search algorithm with secondary search and rollback for improved efficiency.
• Integrated You Only Look Once (version 8) for detection, with a confusion-aware mechanism using corner detection, feature points, and dictionary matching.
• Utilized depth information for precise target localization.

Main Results:

• The grid-based beetle antennae search algorithm demonstrated superior convergence speed and path length compared to mainstream algorithms.
• The confusion-aware mechanism showed high discriminative ability on a custom interference dataset.
• The overall system achieved significant speed improvements (over 89% in challenging environments, over 233% vs. previous work) and met real-time requirements.

Conclusions:

• The proposed grid-based beetle antennae search algorithm and lightweight multi-target detection system significantly enhance UAV tracking performance.
• The system effectively addresses challenges of limited resources, real-time processing, and target confusion, improving robustness and speed.