Advanced algorithms for UAV tracking of targets exhibiting start-stop and irregular motion

Area of Science:

• Computer Vision
• Robotics
• Control Systems

Background:

• Unmanned aerial vehicles (UAVs) struggle with tracking targets exhibiting abrupt velocity changes, intermittent stops, and nonlinear trajectories.
• Conventional tracking algorithms, assuming constant velocity, are inadequate for dynamic, discontinuous motion scenarios common in real-world applications.

Purpose of the Study:

• To develop a robust mathematical framework for UAV tracking that overcomes limitations of traditional methods when dealing with complex target motion.
• To enhance tracking accuracy, continuity, and recovery time for UAVs operating in challenging environments.

Main Methods:

• An adaptive hybrid framework utilizing innovation-based confidence metrics to automatically switch between motion models.
• Enhanced alpha-beta-gamma-delta filtering with jerk compensation for irregular motion.
• SMART-TRACK's 3D-to-2D uncertainty propagation for rapid recovery.
• Flow-guided margin loss to address the motion long-tailed problem.

Main Results:

• Achieved 56.1% Higher Order Tracking Accuracy (HOTA), a 65% improvement over traditional Kalman Filter approaches.
• Innovation-based model switching demonstrated 89.3% accuracy in motion transition detection.
• Enhanced filtering improved irregular motion tracking by 15-25%; SMART-TRACK reduced recovery time from 5.8 to 2.3 seconds.
• Flow-guided margin loss improved large motion tracking by 18.7%; maintained 52.3% average accuracy under environmental corruptions.

Conclusions:

• The proposed adaptive hybrid framework significantly enhances UAV tracking performance, particularly for targets with unpredictable motion.
• The study provides practical guidance for deploying robust UAV tracking systems capable of handling real-world complexities.
• The innovations offer substantial improvements in accuracy, continuity, and recovery speed compared to conventional algorithms.