Reported HOTA
80.37
Overview
KRAfT focuses on UAV swarm tracking under missing observations, noisy detections, and ambiguous associations. It combines formation-aware reasoning with Kalman residual refinement and conservative recovery.
Problem Statement
Tracking multiple UAVs is challenging when targets are small, visually similar, and intermittently missing due to occlusion, detector failure, or motion blur.
Motivation
Pure detection-by-tracking systems can break trajectories when observations are missing or ambiguous. KRAfT improves temporal association by using motion consistency and formation-level cues.
Methodology
The method uses detection input, Kalman prediction, residual-aware refinement, formation-aware association, conservative recovery, and track-continuity reasoning.
Key Contributions
- Formation-aware temporal association for UAV swarm tracking.
- Kalman residual refinement for motion-consistent trajectory updates.
- Conservative recovery strategy for missing observations.
- Robust tracking design for visually small and ambiguous aerial targets.
Experimental Setup
Input modality: vision-based UAV swarm detections and video-derived observations. Task: multi-object UAV swarm tracking with missing-observation recovery.
Quantitative Results
Reported HOTA
70.59
Reported HOTA
65.31
Scores are shown as reported across three evaluation subsets/settings; exact subset names are not restated here.
Demo / Media
My Role
My contributions include method design, tracking pipeline implementation, experiments, result analysis, visualization, and manuscript preparation.
Related Publication
M. H. Rahman and S. Madria. "KRAfT: Kalman Residual Diffusion with Formation Awareness for UAV Swarm Tracking." IEEE ICPR, accepted, 2026.